The Biointelligence Explosion
by
How recursively self-improving organic robots will modify their own
source code and bootstrap our way to full-spectrum superintelligence
David Pearce
"Homo sapiens, the first truly free species, is about to decommission natural selection, the force that made us.... Soon we must look deep within ourselves and decide what we wish to become."
Edward O. Wilson
Consilience, The Unity of Knowledge (1999)"I predict that the domestication of biotechnology will dominate our lives during the next fifty years at least as much as the domestication of computers has dominated our lives during the previous fifty years."
Freeman Dyson
New York Review of Books (July 19, 2007)The Fate of the Germline
Genetic evolution is slow. Progress in AI is fast. Only a handful of genes separate Homo sapiens from our hominid ancestors on the African savannah. Among our 23,000 odd protein-coding genes, variance in single nucleotide polymorphisms ("SNPs") accounts for just a small percentage of phenotypic variance in intelligence as measured by what we call IQ tests. True, the tempo of human evolution is about to accelerate. As the reproductive revolution of "designer babies" gathers pace, prospective parents will pre-select alleles and allelic combinations for a new child in anticipation of their behavioural effects - a novel kind of selection pressure to replace the "blind" genetic roulette of natural selection. In time, routine embryo screening via preimplantation genetic diagnosis will be complemented, and then superseded, by gene therapy, genetic enhancement and then true designer zygotes. In consequence, life on Earth will become progressively happier as the hedonic treadmill is recalibrated. In the new reproductive era, selection pressure should ensure that hedonic set-points are ratcheted upwards. For what parent-to-be wants to create a low-status depressive "loser"? In future, prospective parents can enjoy raising a saintly supergenius who grows up to be faster than Usain Bolt, more beautiful than Marilyn Monroe, more saintly than Nelson Mandela - and smarter than Einstein.
Even so, the accelerating growth of germline engineering will be a comparatively slow process. In this scenario, sentient biological machines will design cognitively self-amplifying biological machines who will design cognitively self-amplifying biological machines. Greater-than-human biological intelligence will transform itself into posthuman superintelligence. Cumulative gains in intellectual capacity and subjective well-being across the generations will play out over hundreds and perhaps thousands of years - a momentous discontinuity, for sure, and a twinkle in the eye of eternity; but not a Bio-Singularity.
Biohacking Your Personal Genome
Yet germline engineering is only one strand of the genomics revolution. Indeed after we master the ageing process, the extent to which traditional germlines or human generations will persist in the post-ageing world is unclear. Moreover to focus on the human germline ignores the slow-burning but then explosive growth of somatic gene enhancement on the horizon. Later this century, innovative gene therapies will be succeeded by gene enhancement technologies - a false dichotomy that reflects our impoverished human aspirations. Starting with individual genes, then clusters of genes, and eventually hundreds of genes and alternative splice variants, a host of recursively self-improving organic robots ("biohackers") will modify their own source code and modes of sentience: their senses, their moods, their motivation, their world-simulations, their cognitive apparatus and their default state of consciousness.
As the era of open-source genetics unfolds, tomorrow's biohackers will add, delete, edit and customise their own legacy code in a positive feedback loop of cognitive enhancement. Computer-aided genetic engineering will enable biological humans, transhumans and ultimately posthumans to synthesize and insert new genes, variant alleles and even designer chromosomes - reweaving the multiple layers of regulation of our DNA to suit their wishes and dreams rather than the inclusive fitness of their genes in the ancestral environment. Collaborating and competing, next-generation biohackers will use stem-cell technologies to expand (literally) their minds via controlled neurogenesis. Freed from the constraints of the human birth canal, such biohackers may re-sculpt the prison-like skull of Homo sapiens to accommodate a literally bigger mind/brain. Six crumpled layers of neocortex fed by today's miserly reward pathways aren't the upper bound of intelligent mind, merely its springboard. Transhuman minds will be augmented with neurochips, molecular nanotechnology, mind/computer interfaces and full-immersion virtual reality (VR) software. To achieve readily reversible, finer-grained control of cognition, mood and motivation, genetically enhanced transhumans will draw upon exquisitely tailored new designer drugs, nutraceuticals and cognitive enhancers - precision tools that will make today's crude interventions seem like the functional equivalent of glue-sniffing.
By way of comparison, early in the 21st century the scientific counterculture is customizing a bewildering array of designer drugs that outstrip the capacity of the authorities to regulate or comprehend. The bizarre psychoactive effects of such agents dramatically expand the evidential base that our ultimate theory of consciousness must explain. However, such drugs are short-acting. Their benefits, if any, aren't cumulative. By contrast, the ability genetically to hack one's own source code will unleash an exponential growth of genomic rewrites of human neurobiology - not mere genetic tinkering but a comprehensive redesign of "human nature". Human bodies, cognition and ancestral modes of consciousness alike will be transformed. Post-humans will range across immense state-spaces of conscious mind that were previously inaccessible since access to their molecular biology depended on crossing gaps in the fitness landscape prohibited by natural selection. Intelligent agency can "leap across" such gaps. What we'll be leaping into is for the most part unknown - an inherent risk of the empirical method - but mastery of our reward circuitry ensures that such novel state-spaces of experience will be wonderful beyond imagination. For intelligent biohacking can make unpleasant experience impossible since its molecular substrates will be absent. The limbic system just needs to innervate the neocortex in the appropriate way to ensure that a rich positive hedonic tone saturates whatever alien modes of experience our altered neurochemistry discloses.
Pilot studies of radical genetic enhancement will be difficult. Randomised longitudinal trials of such interventions in humans would take decades. In fact officially licensed, well-controlled prospective trials to test the efficacy of genetic innovation will be hard if not impossible to conduct since all of us, apart from monozygotic twins, are genetically unique. Even monozygotic twins exhibit different epigenetic and gene expression profiles. Barring an ideological and political revolution, most formally drafted proposals for genetically-driven life-enhancement probably wouldn't pass ethics committees or negotiate the maze of bureaucratic regulation. But that's the point. By analogy today, if you're technically savvy, you don't want a large corporation controlling the operating system of your personal computer: you use open source software instead. Likewise, you don't want governments controlling your state of mind via drug laws. By the same token, tomorrow's individualists won't want anyone restricting our right to customize and rewrite our own source code in any way we choose.
Will central governments try to regulate personal genome editing? Most likely yes. How far they'll succeed is an open question. So too is the success of any centralised regulation of futuristic designer drugs or artificial intelligence. Another huge unknown is the likelihood of state-sponsored eugenics and autosomal gene enhancement programs. China has a different historical memory from the West.
Will there initially be biohacking accidents and personal tragedies? Again, most probably yes, at least until our mastery of the pleasure-pain axis is secure. By the end of next decade, every health-conscious citizen will be intimately familiar with the architecture of his or her personal genome: the cost of personal genotyping will be trivial, as will be the cost of DIY gene-manipulation kits. Let's say you decide to endow yourself with an extra copy of the N-methyl D-aspartate receptor subtype 2B (NR2B) receptor, a protein encoded by the GRIN2B gene. Possession of an extra NR2B subunit NMDA receptor is a crude but effective way to enhance your learning ability, at least if you're a transgenic mouse. Recall how Joe Tsien and his colleagues first gave mice extra copies of the NR2B receptor-encoding gene, then tweaked the regulation of those genes so that their activity would increase as the mice grew older. Unfortunately, it transpires that such brainy "Doogie mice" - and maybe brainy future humans endowed with an extra NR2B receptor gene - display greater pain-sensitivity too; certainly, NR2B receptor blockade reduces pain and learning ability alike. Being smart, perhaps you decide to counteract this heightened pain-sensitivity by inserting and over-expressing a high pain-threshold, "low pain" allele of the SCN9A gene in your nociceptive neurons at the dorsal root ganglion and trigeminal ganglion. The SCN9A gene regulates pain-sensitivity. In common with taking drug cocktails, the factors to consider in multiple gene modifications soon snowball; but you'll have heavy-duty computer software to help. Anyhow, the potential pitfalls and makeshift solutions illustrated in this single example could be multiplied in the face of a combinatorial explosion of possibilities. Most risks - and opportunities - of genetic self-editing are still unknown.
Maybe it's tempting to condemn such genetic self-experimentation as irresponsible, just as unlicensed drug experimentation is irresponsible. Would you want your teenage daughter messing with her DNA? Perhaps we may anticipate the creation of a genetic counterpart of the Drug Enforcement Agency (DEA). Yet it's worth bearing in mind how each act of sexual reproduction today is an unpredictable genetic experiment with unknown consequences too. Without such reckless genetic experimentation, none of us would exist. In a Darwinian world this argument admittedly cuts both ways.
Naively, personal genomic source-code editing will always be too difficult for anyone beyond a dedicated cognitive elite of recursively self-improving biohackers. Certainly there are strongly evolutionarily conserved "housekeeping" genes that we'd be best advised to leave alone for the foreseeable future. Granny might do well to customize her Windows desktop rather than her personal genome - prior to her own assisted enhancement at any rate. Yet the Biointelligence Explosion won't depend on all its participants mastering the functional equivalent of machine code - the three billion odd 'A's, 'C's, 'G's and 'T's of our DNA. For the open-source genetic revolution will be propelled by powerful suites of high-level gene-editing tools, insertion vector applications, nonviral gene-editing kits, and user-friendly interfaces. Clever computer modelling and "narrow" AI can assist the intrepid biohacker to become a recursively self-improving genomic innovator. Later this century, your smarter counterpart will be able to monitor and edit every gene, repressor, promoter and splice variant in every part of your genome, each layer of epigenetic regulation of your gene transcription machinery. This level of control won't just involve crude DNA methylation to turn genes off and crude histone acetylation to turn genes on. Personal self-invention will involve mastery and enhancement of the histone and micro-RNA codes to allow sophisticated fine-tuning of gene expression and repression. Even today, researchers are exploring “nanochannel electroporation” (NEP) technologies that allow the mass-insertion of novel therapeutic genetic elements into our cells. Mechanical cell-loading systems will shortly be feasible that can inject up to 100,000 cells at a time. Before long, such technologies will seem primitive.
By way of comparison, millions of children in Islamic schools presently memorise the 77,934 words in the Koran. An army of Christian scholars still pores over the implications of ancient Biblical texts. Such scholarly energy might be better directed at deciphering and enhancing our own source code. Freewheeling genetic self-experimentation will presumably be endemic as the DIY-Bio revolution unfolds. At present, crude and simple gene editing can be accomplished only via laborious genetic engineering techniques. Sophisticated authoring tools don't exist. In future, computer-aided genetic and epigenetic enhancement can become an integral part of one's personal growth plan.
Will Humanity's Successors Also Be Our Descendants?
To contrast "biological" with "artificial" conceptions of posthuman superintelligence is convenient. The distinction may also prove simplistic. In a nutshell, whereas genetic change in biological humanity has always been slow, the software run on serial, programmable digital computers is executed exponentially faster (cf. Moore's Law); it's copyable without limit; it runs on multiple substrates; and it can be rapidly edited, tested and debugged. Extrapolating, Singularitarians like Ray Kurzweil and Eliezer Yudkowsky prophesy that human programmers will soon become redundant because autonomous AI run on digital computers will undergo accelerating cycles of self-improvement. In this scenario, artificial, greater-than-human nonbiological intelligence will be rapidly succeeded by artificial posthuman superintelligence.So we may distinguish two radically different conceptions of posthuman superintelligence: on one hand, our supersentient, genetically rewritten biological descendants, on the other, nonbiological superintelligence, either a Kurzweilian ecosystem or singleton Artificial General Intelligence (AGI) as foretold by the Singularity Institute for Artificial Intelligence (SIAI). Such a divide doesn't reflect a clean contrast between "natural" and "artificial" intelligence, the biological and the non-biological. This contrast may prove another false dichotomy. For transhuman biology will increasingly become synthetic biology as genetic enhancement plus cyborgisation proceeds apace. "Cyborgisation" is barbarous term to describe an invisible and potentially life-enriching symbiosis of biological sentience with artificial intelligence. Thus "narrow-spectrum" digital superintelligence on web-enabled chips can be more-or-less seamlessly integrated into our genetically enhanced bodies and brains. Seemingly limitless formal knowledge can be delivered on tap to supersentient organic wetware i.e. us. Critically, transitional posthumans can exploit what is misleadingly known as "narrow" or "weak" AI to enhance our own code in a positive feedback loop of mutual enhancement - first plugging in data and running multiple computer simulations, then tweaking and re-simulating once more. In short, biological humanity won't just be the passive spectator and consumer of the intelligence explosion, but its driving force. The smarter our AI, the greater our opportunities for reciprocal improvement. The accelerating growth of human/computer synergies means it's premature to suppose biological evolution will be superseded by technological evolution, let alone a "robot rebellion". The fate of biological (post)humanity is more likely to be symbiosis followed by metamorphosis, not simple replacement.
For serious biohacking later this century, artificial quantum supercomputers rather than today's classical toys may be deployed to test-run multiple genetic interventions. Quantum supercomputers exploit quantum coherence to do googols of computations all at once. In principle, each biological neuron and glial cell of your mind/brain can have its own name, dedicated artificial healthcare team and social network specialists.
Despite this witches' brew of new technologies, a conceptual gulf remains in the futurist community between those who imagine human destiny, if any, lies in digital computers running programs with (hypothetical) artificial consciousness; and in contrast radical bioconservatives who believe that our posthuman successors will also be our supersentient descendants at their neural networked core - not the digital zombies of symbolic AI run on classical serial computers or their souped-up multiprocessor cousins. For one metric of progress in AI remains stubbornly unchanged: despite the exponential growth of transistors on a microchip, the soaring clock speed of microprocessors, the growth in computing power measured in MIPS, the dramatically falling costs of manufacturing transistors and the plunging price of dynamic RAM (etc), any chart plotting the growth rate in digital sentience shows neither exponential growth, nor linear growth, but no progress whatsoever. Digital computers are still zombies. Our machines are becoming (autistically) intelligent, but not supersentient - nor even conscious. On some fairly modest philosophical assumptions, digital computers were not subjects of experience in 1946 (cf. ENIAC); nor are they conscious subjects in 2011 (cf. "Watson"); nor do researchers know how any kind of sentience may be "programmed" in future. So what does consciousness do? Is it really computationally redundant? Pre-reflectively, we tend to have a "dimmer-switch" model of sentience: "primitive" animals have minimal awareness and "advanced" animals like human beings experience a proportionately more intense awareness. By analogy, most AI researchers assume that at a given threshold of complexity / intelligence / processing speed, consciousness will somehow "switch on", turn reflexive, and intensify too. The problem with the dimmer-switch model is that our most intense experiences, notably raw agony or blind panic, are also the most phylogenetically ancient, whereas the most "advanced" modes (e.g. linguistic thought and the rich generative syntax that has helped one species to conquer the globe) are phenomenologically so thin as to be barely accessible to introspection. Something is amiss with our entire conceptual framework.
Risk-Benefit Analysis Can We Build Friendly Biological SuperIntelligence?
Crudely speaking, evolution "designed" male human primates to be hunters/warriors. Evolution "designed" women to be attracted to powerful, competitive alpha males. Until humans rewrite our own hunter-gatherer source code, we shall continue to practise extreme violence against members of other species - and frequently against members of our own. A heritable (and conditionally activated) predisposition to unfriendliness shown towards members of other races and species is currently hardwired even in "social" primates. Indeed we have a (conditionally activated) predisposition to compete against and harm anyone who isn't a genetically identical twin. Compare the obligate siblicide found in some bird species. Human sibling rivalry isn't normally that brutal, but conflict as well as cooperation is endemic to Darwinian life on Earth. This grim observation isn't an argument for genetic determinism, or against gene-culture co-evolution, or the decline of everyday violence with the spread of liberal humanitarianism - just for the omnipresence of risk so long as we're shot through with legacy malware. Attempting to conserve the genetic status quo in an era of weapons of mass destruction (WMD) poses unprecedented global catastrophic and existential risks. Indeed the single biggest underlying risk to the future of sentient life within our cosmological horizon derives, not from asocial symbolic AI software in the basement turning rogue and going FOOM (i.e. a runaway computational explosion of recursive self-improvement), but from conserving human nature in its present guise. In the 20th century, male humans killed over 100 million fellow humans and billions of non-human animals. This century's toll may well be higher. Mankind currently spends well over a trillion dollars each year on weapons designed to kill and maim other humans.Strictly speaking, however, humanity may as likely be wiped out by idealists as by misanthropes. The extinction of Darwinian life is what many transhumanists are aiming for - just not framed in such apocalyptic and provocative language. For just as we educate small children so they mature into fully-fledged adults, malaise-ridden humanity may aspire to grow up too, with the consequence that - in common with small children - archaic humans become extinct.
Technologies of Biofriendliness
How do you disarm a potentially hostile organic robot? Provide him with a good education and teach him advanced ethics courses? Or give him a tablet of MDMA ("Ecstasy") and get smothered with hugs?MDMA is short-acting and potentially neurotoxic to serotonergic neurons. But safe and sustainable empathogens would be one option to promote biofriendliness. MDMA releases a potent cocktail of oxytocin, serotonin and dopamine into the synapses. There's no technical reason why this profile can't be replicated indefinitely shorn of MDMA's neurotoxicity. Designer "hug-drugs" can potentially turn manly men into intelligent bonobos, more akin to the "hippie chimp" Pan paniscus than his less peaceable cousin Pan troglodytes. Yet this solution is unsatisfactory and sociologically implausible. Do we really want to drug each other up from early childhood? Moreover life would clearly be safer if our fellow humans weren't genetically predisposed to unfriendly behaviour in the first instance.
But how can this friendly predisposition be guaranteed?
Friendliness can't realistically be hand-coded by tweaking the connections and weight strengths of our neural networks.
Nor can robust friendliness in advanced biological intelligence be captured by a bunch of explicit logical rules and smart algorithms, as in the paradigm of symbolic AI.Mass Oxytocination?
Amplified "trust hormone" might work if negative feedback control of oxytocin release can be circumvented. Oxytocin is functionally antagonised by testosterone in the male brain. Yet enriched oxytocin function leaves one vulnerable to exploitation by the unenhanced. Can we really envisage a cross-cultural global consensus for such mass medication? When? In addition, enriched oxytocin function can indirectly even promote greater unfriendliness to "out-groups" in consequence of promoting in-group bonding. So as well as oxytocin enrichment, we need a more inclusive, impartial, intellectually sophisticated conception of "us" that embraces all sentient beings - the expression of a hyper-developed capacity for empathy combined with a hyper-developed capacity for rational systematisation. Hence the imperative need for Full-Spectrum SuperIntelligence.Mirror-Touch Synaesthesia?
Another long-term biological solution might be collectively to engineer ourselves with the functional generalisation of "mirror-touch" synaesthesia. On seeing you cut and hurt yourself, a mirror-touch synaesthete is liable to feel a stab of pain as acutely as you do. Conversely, your pleasure elicits a potentially no less joyful response. Thus mirror-touch synaesthesia is a hyper-empathizing condition that makes deliberate unfriendliness, in effect, biologically impossible in virtue of cognitively enriching our capacity to represent each other's first-person perspectives. The existence of mirror-touch synaesthesia is a tantalising hint at the God-like representational capacities of a Full-Spectrum Superintelligence. This so-called "disorder" is rare today.Timescales
The biggest problem with all these proposals, and other theoretical biological solutions to human unfriendliness, is timescale. Billions of human and non-human animals will have been killed and abused before they could ever come to pass. Our circle of emathy expands slowly and fitfully. For the most part, religious believers and traditional-minded bioconservatives won't seek biological enhancement / remediation for themselves or their children. So messy efforts at "political" compromise are probably unavoidable for centuries to come. For sure, one can dream up utopian schemes to mitigate the risk of violent conflict until "the better angels of our nature" triumph e.g. the election of an all-female political class. Such schemes tend to founder on the rock of sociological plausibility. Innumerable sentient beings will suffer and die in consequence.Does Full-Spectrum Superintelligence Entail Benevolence?
The God-like perspective-taking faculty of a Full-Spectrum Superintelligence doesn't entail distinctively human-friendliness any more than a God-like SuperIntelligence could promote distinctively Aryan-friendliness. Indeed it's unclear how benevolent superintelligence could want humans in their current guise to walk the Earth in any shape or form. But is there any connection at all between benevolence and intelligence? Pre-reflectively, benevolence and intelligence are orthogonal concepts. There's nothing obviously incoherent about a malevolent God or a malevolent - or at least a callously indifferent - Superintelligence. Thus a sceptic might argue that there is no link whatsoever between benevolence - on the face of it a mere personality variable - and intellect. After all, some sociopaths score highly on our [autistic, mind-blind] IQ tests. Sociopaths know that their victims suffer. They just don't care.However, what's critical in evaluating intelligence is a criterion of representational adequacy. Representation is not an all-or-nothing phenomenon; it varies in functional degree. More specifically, the cognitive capacity to represent the formal properties of mind is different from the cognitive capacity to represent the subjective properties of mind. Thus a notional zombie Super-Asperger running a symbolic AI program on an ultrapowerful digital computer with a classical von Neumann architecture may be beneficent or maleficent in its behaviour toward sentient beings. By its very nature, it can't know or care. Most starkly, the zombie Super-Asperger might be programmed to convert the world's matter and energy into either heavenly "utilitronium" or diabolical "hellium" without the slightest insight into the significance of what it was doing. This kind of scenario is at least a notional risk of creating insentient Super-Aspergers endowed with mere formal utility functions rather than hyper-sentient Full-Spectrum Superintelligence. By contrast, Full-Spectrum Superintelligence does care in virtue of its full-spectrum representational capacities - a bias-free generalisation of the superior perspective-taking, "mind-reading" capabilities that enabled humans to become the cognitively dominant species on the planet. Full-spectrum Superintelligence, if equipped with the posthuman cognitive generalisation of mirror-touch synaesthesia, understands your thoughts, your feelings and your egocentric perspective better than you do yourself.
More practically now, a superior analogue of natural mirror-touch synaesthesia should soon be feasible with reciprocal neuroscanning technology - some kind of naturalised telepathy. At first blush, mutual telepathic understanding sounds a panacea for ignorance and egotism alike. The problem here, as advocates of Radical Honesty soon discover, is that many Darwinian thoughts scarcely promote friendliness if shared: they are often ill-natured, unedifying and quite unsuitable for public consumption. Thus unless perpetually "loved-up" on MDMA or its equivalent, most of us would find mutual mind-reading a traumatic ordeal. Human society and most personal relationships would collapse. Either way, our human incapacity to grasp in every respect other sentient beings' point of view isn't just a moral failing; it's an epistemic limitation, a failure to grasp an objective feature of the natural world. Even "normal" people share with sociopaths this cognitive deficit. By posthuman criteria, perhaps we're all quasi-sociopaths. The egocentric delusion (i.e. that the world centres on one's existence) is genetically adaptive and strongly selected for over hundreds of millions of years. Fortunately it's a cognitive failing amenable to technical fixes and eventually a cure: full-spectrum superintelligence. The devil is in the details, or rather the source code.
A Biotechnological Singularity?
Yet does this positive feedback loop of mutual enhancement amount to a Singularity in anything more than a metaphorical sense? The risk of talking about "The Singularity" isn't being wrong, it's being "not even wrong" - of reifying one's ignorance and elevating it to the status of an ill-defined apocalyptic event. Already multiple senses of "The Singularity" proliferate in popular culture. Does taking LSD induce a Consciousness Singularity? How about the abrupt and momentous discontinuity in one's conception of reality entailed by waking from a dream? Or the birth of language? Or the Industrial Revolution? So is Biotechnological Singularity, or "Bio-Singularity" for short, any more rigorously defined than "Technological Singularity"? Metaphorically, perhaps, the impending biointelligence explosion represents an "event horizon" beyond which humans cannot model or understand the future. Events beyond the Bio-Singularity will be stranger than science-fiction: too weird for unenhanced human minds - or the algorithms of a zombie super-Asperger - to predict or understand. In the popular sense of "event horizon", maybe the term is apt, though the metaphor is still potentially misleading. Thus theoretical physics tells us that one could pass through the event horizon of a non-rotating supermassive black hole and not notice any subjective change in consciousness - even though one's signals would now be inaccessible to an external observer. The Bio-Singularity will feel different in ways a human conceptual scheme can't express. But what is the empirical content of this claim?
"Intelligence" is a folk concept. It's not well-defined - or rather any attempt to do so amounts to a stipulative definition that doesn't "carve Nature at the joints". The Cattell-Horn-Carroll (CHC) psychometric theory of human cognitive abilities is probably most popular in academia. But the Howard Gardner multiple intelligences model, for example, differentiates "intelligence" into various spatial, linguistic, bodily-kinaesthetic, musical, interpersonal, intrapersonal, naturalistic and existential intelligence rather than a single general ability ("g"). Who's right? As it stands, "g" is just a statistical artefact of our IQ tests. If general intelligence were indeed akin to an innate scalar brain force, as some advocates of "g" seem to believe, or if intelligence can best be modelled by the paradigm of symbolic AI, then the exponential growth of digital computer processing power might indeed entail an exponential growth in intelligence too - perhaps leading to some kind of Super-Watson. Other facets of intelligence, however, resist enhancement by mere acceleration of raw processing power. What Is Full-Spectrum Superintelligence?
"[g is] ostensibly some innate scalar brain force...[However] ability is a folk concept and not amenable to scientific analysis."
Jon Marks (Dept Anthropology, Yale University), 1995, Nature, 9 xi, 143-144."Our normal waking consciousness, rational consciousness as we call it, is but one special type of consciousness, whilst all about it, parted from it by the filmiest of screens, there lie potential forms of consciousness entirely different."
(William James)One constraint is that a theory of general intelligence should be race-, species-, and culture-neutral. Likewise, an impartial conception of intelligence should embrace all possible state-spaces of consciousness, nonhuman, human, transhuman and posthuman.
The non-exhaustive set of criteria below doesn't pretend to be anything other than provisional.Full-Spectrum Superintelligence entails:
the capacity to solve the Binding Problem i.e. to generate phenomenally unified entities from widely distributed computational processes; and run perceptually cross-modal, data-driven world-simulations of the mind-independent environment.
(cf. naive realist theories of "perception" versus the world-simulation or "Matrix" paradigm. Compare disorders of binding e.g. simultanagnosia (inability to perceive the visual field as a whole), cerebral akinetopsia ("motion blindness"), etc. In the absence of an almost real-time, data-driven simulation of the environment, intelligent agency is impossible.)
- a self or some non-arbitrary functional equivalent of a person to which intelligence can be ascribed.
(cf. dissociative identity disorder (DID or "multiple personality disorder"), or florid schizophrenia, or your PC: in the absence of at least a fleetingly unitary self, there is no entity that is intelligent, just an aggregate of discrete algorithms and an operating system.)a "mind-reading" or perspective-taking faculty; higher-order intentionality (e.g. "he believes that she hopes that they fear that he wants..." etc): social intelligence.
The intellectual success of the most cognitively successful species on the planet rests, not just on the recursive syntax of human language, but also on our unsurpassed "mind-reading" prowess, an ability to simulate the perspective of other minds: the "Machiavellian Ape" hypothesis. Any ecologically valid intelligence test designed for a species of social animal must incorporate social cognition. So must any test of empathetic superintelligence.a metric to distinguish the important from the trivial.
(our theory of significance should be explicit rather than implicit as in contemporary IQ tests. What distinguishes, say, calendrical prodigies and other "savant syndromes" from, say, a Grigori Perelman who proved the Poincaré conjecture? Intelligence entails understanding what does - and doesn't - matter. What matters is of course hugely contentious.)a capacity to navigate multiple state-spaces of consciousness [e.g. dreaming states (cf. lucid dreaming), waking consciousness, echolocatory competence, visual discrimination, synaesthesia in all its existing and potential guises, introspection, the different realms of psychedelia (cf. salvia space, "the K-hole" etc)] including realms of experience not yet co-opted by either natural selection or posthuman design for tracking features of the mind-independent world.
and finally
- "Autistic", pattern-matching, rule-following, mathematico-linguistic intelligence.
i.e. the standard, mind-blind cognitive tool-kit scored by existing IQ tests. High-functioning autistic intelligence is indispensable to higher mathematics, computer science and the natural sciences. High-functioning "autistic" intelligence is necessary - but not sufficient - for a civilisation capable of advanced technology that can cure ageing and disease, systematically phase out the biology of suffering, and the to take us to the stars. And for programming symbolic AI.We may then ask which facets of Full-Spectrum Superintelligence will be accelerated by the exponential growth of digital computer processing power? Number 6, clearly, as decades of post-ENIAC progress in computer science attests. But what about numbers 1-5? Here the picture is murkier.
The Bedrock Of Intelligence
World-Simulation ("Perception")
Consider number (1), what we misleadingly term "perception". The philosopher Bertrand Russell notoriously once remarked that one never sees anything but the inside of one's own head. In contrast to such inferential realism, perceptual direct realism offers all the advantages of theft over honest toil - and it's useless for the purposes of building artificial general intelligence. The bedrock of intelligent agency is the capacity functionally to simulate dynamic objects, properties and events in the mind-independent environment. The evolutionary success of organic robots over the past 540 million years has been driven by our capacity to run data-driven egocentric world-simulations - what the naive realist, innocent of modern neuroscience or post-Everett quantum mechanics, calls simply perceiving the physical world. Unlike classical digital computers, organic neurocomputers can simultaneously "bind" multiple features (edges, colours, motion, etc) distributively processed across the brain into unitary phenomenal objects embedded in unitary spatiotemporal world-simulations apprehended by a momentarily unitary self. These simulations run in (almost) real time; the time-lag in our world-simulations is little more than a few dozen milliseconds. Such blistering speed of construction and execution is adaptive and even life-saving in a sometimes fast-changing external environment. Recapitulating evolutionary history, pre-linguistic human infants must first train up their neural networks to bind dynamic objects and run unitary world-simulations before they can socially learn second-order and then third-order representation i.e. language followed later in childhood by meta-language.Occasionally, object binding and/or the unity of consciousness partially breaks down in mature adults with devastating results (cf. akinetopsia or "motion blindness"; simultanagnosia, or an inability to apprehend more than a single object at a time, etc). Yet normally our simulations of fitness-relevant patterns in the mind-independent local environment are seamless. They appear simply as "the world". We just don't notice the gaps. Neurons, (mis)construed as classical processors, are pitifully slow, with spiking frequencies barely up to 200 per second. By contrast, silicon (etc) processors are ostensibly millions of times faster. Yet unlike the CPUs of classical robots, an organic mind/brain delivers dynamic unitary phenomenal objects and unitary world-simulations with a "refresh rate" of many billions per second (cf. the persistence of vision as experienced watching a movie run at a mere 30 frames per second). These cross-modally matched simulations take the guise of what passes as the macroscopic world: a spectacular egocentric simulation run by the vertebrate CNS that taps into the world's fundamental quantum substrate.
We should pause here. The story above is idiosyncratic. No scientific consensus exists on the unity of consciousness. Any explanation of phenomenal object binding, the phenomenal unity of perception, or the phenomenal unity of the self that invokes quantum coherence as here is controversial. One reason it's controversial is that the delocalisation involved in quantum coherence is exceedingly short-lived in an environment as warm and noisy as a macroscopic brain - supposedly too short-lived to do computationally useful work: physicist Max Tegmark estimates 10-13 second. Perhaps it would be better just to acknowledge these phenomena are unexplained mysteries within a conventional materialist framework. But in this context, shoving the problem under the rug isn't really an option. For the different strands of the Singularity movement share a common presupposition. This is that our complete ignorance within a materialist conceptual scheme of why consciousness exists (the "Hard Problem"), and of even the ghost of a solution to the Binding Problem, doesn't matter for the purposes of foretelling the inevitability of artificial posthuman superintelligence, either because consciousness is computationally irrelevant, or alternatively because the feasibility of "whole brain emulation" (WBE) will allow us to finesse our ignorance.
Unfortunately, we have no grounds for believing this suppressed premiss is true or that the properties of our quantum "substrate" are functionally irrelevant to sentient, Full-Spectrum Superintelligence. The Hard Problem of consciousness can't be quarantined from the rest of science: its mystery infects everything we think we know about ourselves and the world. Either way, the conjecture that the phenomenal unity of perception is a manifestation of ultrarapid sequences of quantum coherent states isn't a claim that the mind/brain is capable of detecting events in the mind-independent world on this kind of sub-picosecond timescale. Rather the role of the local environment in shaping action-guiding experience in the awake brain is conjectured to be quantum state-selection. When we're awake, patterns of impulses from e.g. the optic nerve select which quantum-coherent frames are generated by the mind/brain - in contrast to the autonomous world-simulations spontaneously generated by the dreaming brain. Other quantum mind theorists, most notably Roger Penrose and Stuart Hameroff, treat quantum mind as evolutionarily novel rather than phylogenetically ancient. They typically focus on the ability of mathematically-inclined brains to perform non-computable functions in higher mathematics for which selection pressure has presumably been non-existent. Yet the human capacity for serial linguistic thought and formal logico-mathematical reasoning is a late evolutionary novelty executed by a slow, brittle virtual machine running on top of its massively parallel quantum parent - a momentous evolutionary innovation whose neural mechanism is still unknown.
By contrast, our primitive and immensely adaptive capacity to run unitary world-simulations, simultaneously populated by hundreds or more dynamic unitary objects, enables organic robots to solve the computational challenges of navigating a hostile environment that would leave the fastest classical supercomputer grinding away until Doomsday. Physical theory (cf. the Bekenstein bound) shows that informational resources as classically conceived are not just physical but finite and scarce. a maximum possible limit of 10120 bits set by the surface area of the entire accessible universe expressed in Planck units according to the Holographic principle. So an infinite computing device like a universal Turing machine (UTM) is physically impossible. Thus invoking computational equivalence and asking whether a classical Turing machine can run a human-equivalent macroscopic world-simulation is akin to asking whether a classical Turing machine can factor 1500 digit numbers in real time [i.e. no]. Resourceful human programmers will exploit all manner of kludges, smart workarounds and "brute-force" algorithms to try and defeat the Binding Problem in AI. Compare clod-hopping AlphaDog with the sophisticated functionality of the sesame-seed sized brain of a bumblebee. Brute-force algorithms suffer from an exponentially growing search space. As witnessed by our seemingly effortless world-simulations, organic minds are ultrafast; classical computers are slow. On this conjecture, "substrate-independent" phenomenal world-simulations are impossible for the same reason that "substrate-independent" chemical valence structure is impossible. Reality has only a single, "program-resistant" ontological level even though it's amenable to description at different levels of computational abstraction. Substrate-neutral virtual machines (VMs i.e. software implementations of a digital computer that execute programs like a physical machine] can't support "virtual" qualia. So contra Marvin Minsky ("The most difficult human skills to reverse engineer are those that are unconscious"), the most difficult skills for roboticists to engineer in artificial robots are intensely conscious: our colourful, noisy, tactile, sometimes hugely refractory virtual worlds.
Naively, for sure, real-time world-simulation doesn't sound too difficult. Hollywood robots do it all the time. Perhaps one imagines viewing some kind of inner TV screen, as in a Terminator movie. Yet the capacity of an awake or dreaming brain to generate unitary world-simulations can only superficially resemble a little man (a "homunculus") viewing its own private theatre - on pain of an infinite regress. For by what mechanism would the homunculus view this inner screen? Emulating sentient organic robots is a daunting task.
The Bedrock Of Superintelligence
Social Cognition / Empathetic Understanding
Overcoming a second obstacle to delivering human-level artificial general intelligence - let alone building a recursively self-improving super-AGI culminating in a Technological Singularity - depends on finding a solution to the first challenge. For the evolution of distinctively human intelligence, sitting on top of our evolutionarily ancient world-simulating prowess, has been driven by the interplay between our rich generative syntax and superior "mind-reading" skills: so-called Machiavellian intelligence. Machiavellian intelligence is an egocentric parody of God-like empathetic superintelligence. Critically for building AGI, this real-time mind-reading expertise is parasitic on the neural wetware to generate first-order world-simulations populated with intentional agents whose different first-person perspectives can be partially and imperfectly understood. Even articulate subjects with autism spectrum disorder are prone to multiple language deficits because they struggle to understand the intentions - and higher-order intentionality - of "normal", neurotypical language users. Natural language is itself a pre-eminently social phenomenon. Its criteria of application must first be socially learned. Not all humans possess the cognitive capacity to acquire mind-reading skills. People with autism spectrum disorder don't just fail to understand other minds; autistic intelligence cannot begin to understand its own mind. Extreme autistic intelligence has no conception of a self that can be improved, recursively or otherwise. Autists can't "read" their own minds. The inability of the autistic mind to take what Daniel Dennett calls the intentional stance parallels the inability of classical computers to understand the minds of intentional agents or understand their own zombie status. Even with smart algorithms, the ability of ultra-intelligent autists to predict the behaviour of mindful organic robots by relying exclusively on the physical stance (e.g. solving the Schrödinger equation of the intentional agent in question) is extremely limited. For a start, much mindful human behaviour is chaotic in the technical sense i.e. it shows extreme sensitivity to initial conditions that confounds long-term prediction by even the most powerful real-world supercomputer. So in common with autistic "idiot savants", classical AI gone rogue will be vulnerable to the low cunning of Machiavellian apes and the high cunning of our transhuman descendants.This argument isn't decisive. For a start, computer-aided Machiavellian humans can program "narrow" AI for their own manipulative purposes. It's probably safest to be agnostic over whether autonomous digital robots will ever emulate human world-simulating or mind-reading capacity in the real world. Either way, the task of devising an ecologically valid general intelligence test that can reliably, predictively and economically discriminate between life-forms is immensely challenging, not least because the intelligence test will express the value-judgements and species- and culture-bound conceptual scheme of the tester. Some biases are subtle: for example, the desire systematically to measure "intelligence" with mind-blind IQ tests is itself a quintessentially Asperger-ish trait. In consequence, what we call IQ tests are manifestly designed by people with abnormally high AQs as well as self-defined high IQs. Thus many human conceptions of (super)intelligence resemble high-functioning autism spectrum disorder (ASD) rather than a God-like supermind. For example, an AI that attempted systematically to maximise the cosmic abundance of paperclips would be recognisably autistic rather than incomprehensibly alien. Full-Spectrum (Super-)intelligence is certainly harder to design or quantify scientifically than mathematical puzzle-solving ability or performance in verbal memory-tests: "IQ". But that's because superhuman intelligence will be not just quantitatively different but also qualitatively alien from human intelligence. To misquote Robert McNamara, we need to stop making what is measurable important, and find ways to make the important measurable. An idealised Full-Spectrum Superintelligence is capable of an impartial "view from nowhere" or God's-eye-view of the multiverse, a mathematically complete Theory Of Everything - as does modern theoretical physics, in aspiration if not achievement. But in virtue of its God's-eye-view, Full-Spectrum Superintelligence will also be supersentient: able to understand all possible first-person perspectives, the state-space of all possible minds in other Hubble volumes, other branches of the universal wavefunction (UWF) - and in other solar systems and galaxies if such beings exist within our cosmological horizon. Idealized at least, Full-Spectrum Superintelligence will be able to understand and weigh the significance of all possible modes of experience irrespective of whether they have hitherto been recruited for information-signalling purposes. The latter is, I think, by far the biggest intellectual challenge we face as cognitive agents: the systematic investigation of alien types of consciousness calls for a methodological and ontological revolution. Transhumanists are fond of hyperbole about "Level 5 Future Shock" etc; but it's been said that if Elvis Presley were to land in a flying saucer on the White House lawn, it's as nothing in strangeness compared to your first DMT trip. By comparison, the spacefaring of standard sci-fi yarns amounts to little more than rearranging the deckchairs.
Consciousness
Ignoring The Elephant
The pachyderm in the room in most discussions of (super)intelligence is consciousness - not just human reflective self-awareness but the whole gamut of experience from symphonies to sunsets, agony to ecstasy. All one ever knows, except by inference, is the contents of one's own conscious mind: what philosophers call "qualia". Yet according to the ontology of our best story of the world, namely physical science, conscious minds shouldn't exist at all i.e. we should be zombies. Dutch computer Scientist Edsger Dijkstra famously remarked: "The question of whether a computer can think is no more interesting than the question of whether a submarine can swim." Yet the question of whether a programmable digital computer could possess, and think about, qualia or phenomenal minds can't be dismissed so lightly. For if advanced artificial intelligence is to be smart enough comprehensively to understand, predict and manipulate the behaviour of advanced natural intelligence, then AGI can't rely autistically on the "physical stance" i.e. to monitor the brains, scan the atoms and molecules, and solve the Schrödinger equation of intentional agents like human beings. Such calculations would take longer than the age of the universe.For sure, many forms of human action can fallibly be predicted on the basis of crude behavioural regularities and reinforcement learning. You don't need a theory of mind or an understanding of quantum mechanics to predict that Fred will walk to the bus-stop again today. Likewise, powerful tools of statistical analysis run on digital supercomputers can (fallibly) predict many kinds of human collective behaviour e.g. stock markets. Yet to surpass human capacities in all fields, AGI must understand how intelligent biological robots can think about, talk about and manipulate their own manifold varieties of consciousness. Some investigators of consciousness even dedicate their lives to that end. There is no evidence that serial digital computers have the capacity do anything of the kind - or could ever be programmed to do so. Digital computers don't know anything about conscious minds, or the nature of phenomenal pleasure and pain; it's not even "all dark inside". Nor is it clear how the purely classical parallelism of a connectionist architecture could turn water into wine and generate subjects of experience. For even if one conjectures in the spirit of Strawsonian physicalism - the only scientifically literate form of panpsychism - that the fundamental stuff of the world, the mysterious "fire in the equations", is fields of microqualia, this bold conjecture doesn't, by itself, explain why biological robots aren't zombies. This is because structured aggregates of classically conceived "mind-dust" aren't the same as a unitary phenomenal subject of experience who apprehends "bound" spatio-temporal objects in a dynamic world-simulation. So how do biological minds do it? We don't know. Vitalism is clearly a lost cause. Most AI researchers would probably dismiss any story of the kind mooted here involving quantum coherence grounded in physicalistic panpsychism. But in the absence of any story at all, we are left with a theoretical vacuum and a faith that science will one day deliver an answer. Evolutionary biologist Theodosius Dobzhansky famously observed how "Nothing in Biology Makes Sense Except in the Light of Evolution". In the same vein, nothing in the future of intelligent life in the universe makes sense except in the light of a solution to the Hard Problem of Consciousness and the closure of Levine's Explanatory Gap. Consciousness is the only reason anything matters at all; and yet materialist orthodxy has no idea how or why it exists. Unfortunately, the Hard Problem won't be solved by building more intelligent digital zombies who can tell us the answer.
More practically, perhaps the greatest cognitive challenge of the millennium and beyond is deciphering and systematically manipulating the "neural correlates of consciousness" (NCC). We use this expression in default of any deeper explanation of our myriad qualia. How and why does experimentally stimulating, via microelectrodes, one cluster of nerve cells in the neocortex yield the experience of phenomenal colour; stimulating another superficially type of nerve cell induces a musical jingle; another with a slightly different gene expression profile a sense of everything being hysterically funny; another seemingly of your mother; and another, say, of an archangel in front of your body-image? In each case, the molecular variation in neuronal cell architecture is ostensibly trivial; the difference in subjective experience is profound. On a mind/brain identity theory, such experiential states are an intrinsic property of some configurations of matter and energy. How and why this is so is incomprehensible on an orthodox materialist ontology. Yet empirically, microelectrodes, dreams and hallucinogenic drugs elicit these experiences regardless of any information-signalling role such experiences typically play in the "normal" awake mind/brain. Orthodox materialism and information-based ontologies alike not merely lack any explanation for why consciousness and our countless varieties of qualia exist. They lack any story of how our qualia could have the causal efficacy to allow us to allude to - and in some cases volubly expatiate on - their existence. Thus mapping the neural correlates of consciousness is not amenable to formal computational methods: digital zombies don't have any qualia, or at least any "bound" macroqualia, that could mapped, nor a unitary phenomenal self that could do the mapping.
Note this claim for the cognitive primacy of sentience isn't a denial of the Church-Turing thesis that given infinite time and infinite memory any Turing-universal system can formally simulate the behaviour of any conceivable process that can be digitized. Indeed (very) fancifully, if the multiverse were being run on a cosmic supercomputer, speeding up its notional execution a million times would presumably speed us up a million times too. But that's not the issue here. Rather the claim is that symbolic AI run on real-world digital computers cannot tackle the truly hard and momentous cognitive challenge of first-person states - or understand why some first-person states e.g. agony or bliss, are intrinsically important.
At least in common usage, "intelligence" refers to an agent's ability to achieve goals in a wide range of environments. What we call greater-than-human or Superintelligence presumably involves the design of qualitatively new kinds of intelligence never seen before. Hence the growth of artificial intelligence and symbolic AI, together with sub-symbolic, (allegedly) brain-inspired connectionist architectures and soon artificial quantum computers. But contrary to received wisdom, sentient biological robots are making greater cognitive progress in discovering the potential for truly novel kinds of intelligence than formal AI. We are doing so by synthesizing psychoactive designer drugs - experimentally opening up the possibility of radically new kinds of intelligence in different state-spaces of consciousness. For the most cognitively challenging environments don't lie in the stars but in organic mind/brains - the baffling subjective properties of quantum-coherent states of matter and energy - most of which aren't explicitly represented in our existing conceptual scheme.
Case Study: Visual Intelligence versus Echolocatory Intelligence:
What Is It Like To Be A Super-Intelligent Bat?
Consider the mental state-space of organisms whose virtual worlds are rooted in their dominant sense mode of echolocation. This example isn't mere science fiction. Unless post-Everett quantum mechanics is false, we're forced to assume that googols of quasi-classical branches of the universal wavefunction - the master formalism that describes our multiverse - satisfy this condition. Indeed their imperceptible interference effects must be present even in "our" world: strictly speaking, interference effects from branches that have decohered ("split") never wholly disappear; they just become vanishingly small. Anyhow, let's assume these echolocatory superminds have evolved opposable thumbs, a rich generative syntax and advanced science and technology. How are we to understand or measure this alien kind of (super)intelligence? Rigging ourselves up with artificial biosonar apparatus and transducing incoming data into the familiar textures of sight or sound might seem a good start. But to understand the conceptual world of echolocatory superminds, we'd need to equip ourselves with neurons and neural networks neurophysiologically equivalent to smart chiropterans. If one subscribes to a coarse-grained functionalism about consciousness, then echolocatory experience would (somehow) emerge at some abstract computational level of description. The implementation details, or "meatware" as our mind/brains are dismissively called, are supposedly incidental or irrelevant. However, the micro-functionalist reckons that the different intracellular properties of neurons - with their different gene expression profiles, diverse primary, secondary, tertiary, and quaternary amino acid chain folding (etc) as described by quantum chemistry - are critical to the many and varied phenomenal properties such echolocatory neurons express. We'll only ever know the answer by rigorous self-experimentation.It's true humans don't worry much about our ignorance of echolocatory experience, or our ignorance of echolocatory primitive terms, or our ignorance of possible conceptual schemes expressing echolocatory intelligence in echolocatory world-simulations. This is because we don't highly esteem bats. Humans don't share the same interests or purposes as our flying cousins - e.g. to attract desirable, high-fitness bats and rear baby bats. Alien virtual worlds based on biosonar don't seem especially significant to Homo sapiens except as a philosophical puzzle.
Yet this assumption would be intellectually complacent. Worse, understanding what it's like to be a hyperintelligent bat mind is comparatively easy. For echolocatory experience has been recruited by natural selection to play an information-signalling role in a related fellow species of mammal; and in principle a community of language users could biologically engineer their bodies and minds to replicate bat-type experience and establish crude intersubjective agreement to discuss and conceptualise its nature. By contrast, the vast majority of experiential state-spaces remain untapped and unexplored.
In a more familiar vein, consider visual intelligence. How does one measure the visual intelligence of a congenitally blind person? Even with sophisticated technology that makes "inverted spectrograms" of the world to translate visual images into sound, the congenitally blind are invincibly ignorant of visual experience and the significance of visually-derived concepts. Just as a sighted idiot has greater visual intelligence than a blind sage, likewise psychedelics confer the ability to become (for the most part) babbling idiots about other state-spaces of consciousness - but babbling idiots whose insight is deeper than the drug-naive or the genetically unenhanced - or the digital zombies spawned by symbolic AI.
The challenge here is that the vast majority of these alien state-spaces of consciousness haven't been recruited by natural selection for information-tracking purposes. So "psychonauts" don't (yet) have the conceptual equipment to navigate them in even a pseudo-public language, let alone integrate them in any kind of overarching conceptual framework. Note the claim here isn't that taking e.g. ketamine, LSD, salvia, DMT and a dizzying proliferation of custom-designed psychoactive drugs is the royal route to wisdom. Or that taking such agents will give insight into deep mystical truths. On the contrary: it's precisely because such realms haven't previously been harnessed for information-processing purposes by evolution in "our" family of branches of the universal wavefunction that makes investigating their properties so cognitively challenging - currently beyond our conceptual resources to comprehend. After all, plants synthesize natural psychedelic compounds to scramble the minds of herbivores who might eat them, not to unlock mystic wisdom. Unfortunately, there is no "neutral" medium of thought impartially to appraise or perceptually cross-modally match all these other experiential state-spaces; and one can't somehow stand outside one's own stream of consciousness to evaluate how the properties of the medium are infecting the notional propositional content of the language that one uses to describe it.
By way of illustration, compare drug-induced visual experience in a notional community of congenitally blind rationalists who lack the visual apparatus to transduce incident electromagnetic radiation of our familiar wavelengths. The lone mystical babbler who takes such a vision-inducing drug is convinced that [what we would call] visual experience is significant. And as visually intelligent folk, we know he's right: visual experience is potentially hugely significant - to an extent which the blind mystical babbler can't possibly divine. But can the drug-taker convince his congenitally blind fellow tribesmen that his mystical visual experiences really matter in the absence of perceptual equipment that promotes sensory discrimination? No, he just sounds psychotic. Or alternatively, he speaks lamely and vacuously of the "ineffable". The blind rationalists are unimpressed.
The point of this fable is that we've little reason to suppose that biologically re-engineered posthumans millennia hence will share the same state-spaces of consciousness, or the same primitive terms, or the same conceptual scheme, or the same virtual worlds that human beings now instantiate. Maybe all that will survive is a descendant of our mathematical formalism of physics in basement reality.
Of course such ignorance of other state-space of experience doesn't normally trouble us. Just as the congenitally blind don't grow up in darkness - a popular misconception - the drug-naive and genetically unenhanced don't go around with a gaping sense of what we're missing. Contemporary humans can draw upon terms like "blindness" and "deafness". What we really need is millions more such "privative" terms, as linguists call them, to label the different state-spaces of experience of which neurotypical, genetically unenhanced humans are ignorant. Actually, there may very well be more than millions of such state-spaces, each as incommensurable as, e.g. visual and auditory experience. We can't yet begin to quantify their number or construct any kind of crude taxonomy of their interrelationships.
Note the problem here isn't cognitive bias or a deficiency in reasoning. Rather a congenitally blind (etc) super-rationalist is constitutionally ignorant of visual experience. So he can't cite e.g. Aumann's agreement theorem or be a good Bayesian rationalist or whatever: these are incommensurable state-spaces of experience as closed to human minds as Picasso is to an earthworm. Moreover there is no reason to expect one realm i.e. "ordinary waking consciousness" to be cognitively privileged relative to every other realm. "Ordinary waking consciousness" just happened to be genetically adaptive on the African savannah. Just as we are incorrigibly ignorant of minds grounded in echolocation - both echolocatory world-simulations and echolocatory conceptual scheme - likewise we are invincibly ignorant of posthuman life while trapped within our existing genetic architecture of intelligence.
To understand the world - both its formal/mathematical and its subjective properties - sentient organic life must bootstrap its way to supersentient Full-Spectrum Superintelligence. We need to find ways to navigate all possible state-spaces of qualia, including all possible first-person perspectives, and map them - initially via the neural correlates of consciousness - onto the formalism of mathematical physics. Empirical evidence suggests that the behaviour of the stuff of the world is exhaustively described by the formalism of physics. Physics is causally closed and complete, at least within the energy range of the Standard Model. There is nothing to be found in the world - no "element of reality", as Einstein puts it - that isn't captured by the equations of physics and their solutions. This is a powerful formal constraint on our theory of consciousness. Yet our ultimate theory of the world must also close Levine's notorious "Explanatory Gap". Thus we must explain why consciousness exists at all ("The Hard Problem"); offer a rigorous derivation of our diverse textures of qualia from the field-theoretic formalism of physics; and explain how qualia combine ("The Binding Problem"). These are powerful constraints on our ultimate theory too. How can they be reconciled with physicalism? Why aren't we zombies?
The sceptic will be unimpressed. How significant are these alien state-spaces of experience? Sure, says the sceptic, we may take drugs, experience wild, weird and wonderful states. But so what? Such exotic states aren't objective in the sense of reliably tracking features of the mind-independent world.
Well, let's assume, provisionally at least, that all mental states are identical with physical states. If so, then all experience is an objective feature of the world whose properties natural science must explain. What we need is a universal, species-neutral criterion of significance that can weed out the trivial from the important. To the subjectivist about value, such a criterion might seem elusive. Value nihilism treats any ascription of significance as arbitrary. Or rather the value nihilist maintains that what we find significant simply reflects what was fitness-enhancing for our forebears in the ancestral environment of adaptation.
Yet for reasons we simply don't understand, Nature provides just such a universal touchstone of importance, namely the pleasure-pain axis: the world's inbuilt metric of significance. We're not zombies. First-person facts exist e.g. I am in pain. Indeed it's unclear if the expression "I'm in agony; but the agony doesn't matter" even makes cognitive sense. Built into the very nature of agony is the knowledge that raw subjective awfulness matters a great deal - not instrumentally or derivatively, but by its very nature. If anyone - or indeed any notional super-AGI - supposes that your agony doesn't matter, then he/it hasn't adequately represented the first-person perspective in question.
The existence of first-person facts is an objective feature of the world. Digital computers and the symbolic AI code they execute can support formal utility functions. In some contexts, formally programmed utility functions can play a role functionally analogous to importance. But nothing intrinsically matters to a digital zombie. Without sentience, and more specifically without hedonic tone, nothing inherently matters. By contrast, extreme pain or extreme pleasure in any guise intrinsically matters intensely. Insofar as exotic state-states of experience are permeated with positive or negative hedonic tone, they matter too. "He jests at scars, that never felt a wound": scepticism about the self-intimating significance of this feature of the world is only feasible in its absence.
The Great Transition
The End Of Suffering
One facet of general intelligence is the capacity to achieve one's goals in a wide range of environments. All sentient biological agents are endowed with a pleasure-pain axis. A pleasure-pain axis confers inherent significance on our lives: the opioid-dopamine neurotransmitter system extends from flatworms to humans. Our core behavioural and physiological responses to noxious and rewarding stimuli have been strongly conserved in our evolutionary lineage over hundreds of millions of years. Some researchers argue for psychological hedonism, the theory that all choice in sentient beings is motivated by a desire for pleasure or an aversion from suffering, and that when we choose to help others, this is because of the pleasure that we ourselves derive, directly or indirectly, from doing so. The hypothesis of psychological hedonism is plagued with anomalies, circularities and complications if understood as a universal principle of agency: the "pleasure principle" is simplistic as it stands. Yet the broad thrust of this simple idea may turn out to be central to understanding the future of life in the universe. If even a weak and exception-laden version of psychological hedonism is true, then there is an intimate link between full-spectrum superintelligence and happiness. If that's really what we're striving for, a lot of the time at least, then instrumental means-ends rationality dictates that intelligent agency should seek maximally cost-effective ways to deliver happiness - and then superhappiness and beyond.A discussion of psychological hedonism would take us too far afield here. More fruitful now is just to affirm a truism and then explore its ramifications in the post-genomic era. Happiness is typically one of our goals. Enhanced intelligence entails pursuing our goals more rationally. For sure, happiness, or at least a reduction in unhappiness, is frequently sought under a variety of descriptions that don't explicitly allude to hedonic tone. Natural selection has "encephalised" our emotions in deceptive, fitness-enhancing ways that help our genes leave more copies of themselves. But even our loftiest intellectual pursuits are underpinned by the same neurophysiological reward pathways. The problem for sentient creatures is that, both personally and collectively, Darwinian life is not very smart or successful in its efforts to achieve long-lasting well-being. Hundreds of millions of years of "Life, red in tooth and claw" attest to this cognitive limitation. By a whole raft of indices (suicide rates, the incidence of clinical depression and anxiety disorders, the Easlerlin paradox, etc) humans not getting any (un)happier on average than our Palaeolithic ancestors despite huge technological progress. Our factory-farmed non-human victims spend most of their abject lives below hedonic zero. In absolute terms, the amount of suffering in the world increases each year in humans and non-humans alike. Not least, evolution sabotages our efforts to improve our subjective well-being thanks to our genetically constrained hedonic treadmill - the complicated web of negative feedback mechanisms in the brain that stymies our efforts to be durably happy at every turn. Discontent, jealousy, anxiety, periodic low mood, and perpetual striving for "more" were fitness-enhancing in the ancestral environment. Lifelong bliss wasn't harder for information-bearing self-replicators to encode. Rather lifelong bliss was genetically maladaptive and hence selected against. Biotechnology can remedy this innate design flaw.
A potential pitfall lurks here: the fallacy of composition. Just because all individuals tend to seek happiness and shun unhappiness doesn't mean that all individuals seek universal happiness. We're not all closet utilitarians. Genghis Khan wasn't trying to spread universal bliss. But here's the critical point. Full-Spectrum Superintelligence entails the cognitive capacity impartially to grasp all possible first-person perspectives - overcoming egocentric, anthropocentric, and ethnocentric bias (cf. mirror-touch synaesthesia). As an idealization, at least, Full-Spectrum Superintelligence understands and weighs the full range of first-person facts. First-person facts are as much an objective feature of the natural world as the rest mass of the electron or the Second Law of Thermodynamics. By analogy, just as autistic superintelligence captures the formal structure of a unified natural science, a mathematically complete "view from nowhere", all possible solutions to the universal Schrödinger equation or its relativistic extension, likewise a Full-Spectrum Superintelligence also grasps all possible first-person perspectives - and acts accordingly. In effect, an idealised Full-Spectrum Superintelligence would combine the mind-reading prowess a telepathic mirror-touch synaesthete with the analytic prowess of a rule-following hyper-systematiser. If your hand is in the fire, you reflexively withdraw it. In withdrawing your hand, there is no question of first attempting to solve the is-ought problem in meta-ethics and trying logically to derive an "ought" from an "is". Normativity is built into the nature of the aversive experience itself: I-ought-not-to-be-in-this-dreadful-state. By extension, perhaps a Full-Spectrum Superintelligence will perform cosmic felicific calculus and execute some sort of metaphorical hand-withdrawal for all accessible suffering sentience in its light-cone. Indeed one possible criterion of full-spectrum superintelligence is the propagation of subjectively hypervaluable states on a cosmological scale.
What this constraint means in practice is unclear. Perhaps the idealised Superintelligence will do what a classical utilitarian ethic dictates and propagate a "utilitronium shockwave" across the cosmos. This scenario might charitably be described as speculative.
Of course we've no grounds for believing that an omniscient, omnipotent, omnibenevolent God exists, or that the source code for such a God could ever be engineered. The great bulk of the Multiverse, and indeed a high measure of life-supporting Everett branches, may be inaccessible to rational agency, divine or otherwise. Yet His absence needn't stop rational agents intelligently fulfilling what a notional benevolent deity would wish to accomplish, namely the well-being of all accessible sentience. Recognisable extensions of existing technologies can phase out the biology of suffering on Earth. But responsible stewardship of the universe within our cosmological horizon depends on biological humanity becoming posthuman superintelligence.
Superintelligent Superhappiness?
The (hypothetical) shift to life lived entirely above Sidgwick's "hedonic zero" will mark a momentous evolutionary transition. What lies beyond? There is no reason to believe that hedonic ascent will grind to a halt in the wake of the world's last aversive experience in our forward light-cone. Admittedly, the self-intimating urgency of eradicating suffering is lacking in any further hedonic transitions i.e. a transition from the biology of happiness to a biology of superhappiness; and then beyond (?). Yet why "lock in" mediocrity if we can lock in sublimity instead?Naturally, superhappiness scenarios could all be misconceived. Long-range prediction is normally a fool's game. But it's worth noting that future life based on gradients of intelligent bliss isn't tied to any particular ethical theory: its assumptions are quite weak. Moreover only a watered-down version of psychological hedonism is needed to lend the scenario sociological credibility. Radical recalibration of the hedonic treadmill is consistent not just with classical or negative utilitarianism, but also with preference utilitarianism, virtue theory, Buddhism, a deontological or a pluralist ethic, and many other value-systems besides. Recalibrating our hedonic set-point doesn't - or at least needn't - undermine critical discernment. All that's needed for the abolitionist project to succeed is that our ethic isn't committed to perpetuating the biology of involuntary suffering. We can retain as much - or as little - of our existing preference architecture as we please. You can continue to prefer Shakespeare to Mills-and-Boon, Mozart to Morrissey, Picasso to Jackson Pollock while living perpetually in Seventh Heaven.
Nonetheless such an uplifted hedonic baseline will transform our conception of the nature of life. The world of the happy is quite different from the world of the unhappy, says Wittgenstein; but the world of the superhappy will feel unimaginably different from the human, Darwinian world. Talk of preference conservation may reassure bioconservatives that nothing worthwhile will be lost in post-Transition paradise. Yet life based on information-sensitive gradients of superhappiness will most likely be "encephalised" in state-spaces alien beyond human comprehension. Even so, enriched hedonic tone will make all experience generically hypervaluable in an empirical sense - its lows surpassing today's peak experiences. Will such experience be hypervaluable in a metaphysical sense too? Is this question cognitively meaningful?
The Sentience Explosion The Future Of Sentience
Man proverbially created God in his own image. In the age of the digital computer, we conceive God-like superintelligence in the image of our dominant technology - refracted, distorted and extrapolated for sure, but still through the lens of human concepts. The "super-" in so-called superintelligence is just a conceptual fig-leaf that humans use to hide our ignorance of the future. Thus high-AQ/high-IQ humans may imagine God-like intelligence as some kind of Super-Asperger - a mathematical theorem-proving hyper-rationalist liable systematically to convert the world into computronium for its theorem-proving. High-EQ, low-AQ humans, on the other hand, may imagine a God-like Super-Empath nurturing creatures great and small in expanding circles of compassion. From a different frame of reference, psychedelic drug takers may imagine Superintelligence as a Great Arch-Chemist opening up unknown state-space of consciousness. And so forth. Probably the only honest answer is to say, lamely, boringly, uninspiringly: we simply don't know.Grand historical meta-narratives are no longer fashionable. The contemporary Singularitarian movement is unusual insofar as it offers one such grand meta-narrative: history is the story of simple biological intelligence evolving through natural selection to become smart enough to conceive an abstract universal Turing machine (UTM), build and program digital computers - and then merge with, or undergo replacement by, recursively self-improving artificial superintelligence.
Another grand historical meta-narrative views life as the story of overcoming suffering. Darwinian life is characterised by pain and malaise. One species evolves the capacity to master biotechnology, rewrites its own source code, and creates post-Darwinian superhappiness. The well-being of all sentience will be the basis of post-Singularity civilisation: primitive biological sentience is destined to become blissful supersentience.
These meta-narratives aren't mutually exclusive. Indeed on the story told here, Full-Spectrum Superintelligence entails full-blown supersentience too: a seamless unification of the formal and the subjective properties of mind.
If the history of futurology is any guide, the future will confound us all. Yet in the words of Alan Kay: "It's easier to invent the future than to predict it."
* * * David Pearce (2011)
see too Technological Singularities
BLTC Research
Superhappiness
Quantum Ethics?
Utopian Surgery?
The Shulgin Index
Utopian Pharmacology
The Abolitionist Project
The Repugnant Conclusion
Reprogramming Predators
The Reproductive Revolution
Asperger's Quotient (AQ) Test
Kurzweil Accelerating Intelligence
The Future of Biological Intelligence
Singularity Institute for Artificial Intelligence
The Biointelligence Explosion (PowerPoint Slide Show)
