The Biointelligence Explosion

Abstract

This essay explores how recursively self-improving organic robots will modify their own genetic source code and bootstrap our way to full-spectrum superintelligence. Starting with individual genes, then clusters of genes, and eventually hundreds of genes and alternative splice variants, tomorrow’s biohackers will exploit “narrow” AI to debug human source code in a positive feedback loop of mutual enhancement. Genetically enriched humans can potentially abolish aging and disease; recalibrate the hedonic treadmill to enjoy gradients of lifelong bliss, and phase out the biology of suffering throughout the living world.

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)

Illah R. Nourbakhsh on Pearce’s “The Biointelligence Explosion”

The Optimism of Discontinuity

In The Biointelligence Explosion, David Pearce launches a new volley in the epic, pitched battle of today’s futurist legions. The question of this age is: machine or man? And neither machine nor man resembles the modern-day variety. According to the Singularity’s version of foreshadowed reality, our successors are nothing like a simulacrum of human intelligence; instead they vault beyond humanity along every dimension, achieving heights of intelligence, empathy, creativity, awareness and immortality that strain the very definitions of these words as they stand today. Whether these super-machines embody our unnatural, disruptive posthuman evolution, displacing and dismissing our organic children, or whether they melt our essences into their circuitry by harvesting our consciousnesses and qualia like so much wheat germ, the core ethic of the machine disciples is that the future will privilege digital machines over carbon-based, analog beings.

Pearce sets up an antihero to the artificial superintelligence scenario, proposing that our wetware will shortly become so well understood, and so completely modifiable, that personal bio-hacking will collapse the very act of procreation into a dizzying tribute to the ego. Instead of producing children as our legacy, we will modify our own selves, leaving natural selection in the dust by changing our personal genetic makeup in the most extremely personal form of creative hacking imaginable. But just like the AI singularitarians, Pearce dreams of a future in which the new and its ancestor are unrecognizably different. Regular humans have depression, poor tolerance for drugs, and, let’s face it, mediocre social, emotional and technical intelligence. Full-Spectrum Superintelligences will have perfect limbic mood control, infinite self-inflicted hijacking of chemical pathways, and so much intelligence as to achieve omniscience bordering on Godliness.

The Singularity proponents have a fundamentalist optimism born, as in all religions, of something that cannot be proven or disproven rationally: faith. In their case, they have undying faith in a future discontinuity, the likes of which the computational world has never seen. After all, as Pearce points out, today’s computers have not shown even a smattering of consciousness, and so the ancestry of the intelligent machine, a machine so fantastically powerful that it can eventually invent the superintelligent machine, is so far an utter no-show. But this is alright if we can believe that with Moore’s Law comes a new golden chalice: a point of no return, when the progress of Artificial Intelligence self-reinforces, finally, and takes off like an airplane breaking ground contact and suddenly shooting upward in the air: a discontinuity that solves all the unsolvable problems. No measurement of AI’s effectiveness before the discontinuity matters from within this world view; the future depends only on the shape of a curve, and eventually all the rules will change when we hit a sudden bend. That a technical sub-field can depend so fully, not on early markers of success, but on the promise of an unknown future disruption, speaks volumes about the discouraging state of Artificial Intelligence today. When the best recent marker of AI, IBM’s Watson, wins peculiarly by responding to a circuit-driven light in 8 ms, obviating the chances of humans who must look at a light and depend on neural pathways orders of magnitude slower, then AI Singularity cannot yet find a machine prophet.

Pearce is also an optimist, presenting an alternative view that extrapolates from the mile marker of yet another discontinuity: when hacker-dom successfully turns its tools inward, open-sourcing and bio-hacking their own selves to create recursively improving bio-hackers that rapidly morph away from mere human and into transcendental Superintelligence. The discontinuity is entirely different from the AI Singularity, and yet it depends just as much on a computational mini-singularity. Computers would need to provide the simulation infrastructure to enable bio-hackers to visualize and test candidate self-modifications. Whole versions of human-YACC and human-VMWare would need to compile and run entire human architectures in dynamic, simulated worlds to see just what behaviour will ensue when Me is replaced by Me-2.0. This demands a level of modelling, analog simulation and systems processing that depend on just as much of a discontinuity as the entire voyage. And then a miracle happens becomes almost cliché when every technical obstacle to be surmounted is not a mountain, but a hyperplane of unknown dimensionality!

But then there is the hairy underbelly of open-source genetics, namely that of systems engineering and open-source programming in general. As systems become more complex, Quality Assurance (QA) becomes oxymoronic because tests fail to exhaustively explore the state-space of possibilities. The Toyota Prius brake failures were not caught by engineers whose very job is to be absolutely sure that brakes never, ever fail, because just the right resonant frequency, combined with a hybrid braking architecture, combined with just the right accelerometer architecture and firmware, can yield a one-in-a million rarity a handful of times, literally. The logistical tail of complexity is a massive headache in the regime of QA, and this bodes poorly for open-sourced hacking of human systems, which dwarf the complexity of Toyota Prius exponentially. IDE’s for bio-hacking; debuggers that can isolate part of your brain so that you can debug a nasty problem without losing consciousness (Game Over!); version control systems and repositories so that, in a panic, you can return your genomic identity to a most recent stable state- all of these tools will be needed, and we will of course be financially enslaved to the corporations that provide these self-modification tools. Will a company, let’s call it HumanSoft, provide a hefty discount on its insertion vector applications if you agree to do some advertising—your compiled genome always drinks Virgil’s Root Beer at parties, espousing its combination of Sweet Birch and Molasses? Will you upgrade to HumanSoft’s newest IDE because it introduces forked compiling—now you can run two mini-me’s in one body and switch between them every 5 s by reprogramming the brain’s neural pathways.

Perhaps most disquieting is the law of unintended consequences, otherwise known as robotic compounding. In the 1980s, roboticists thought that they could build robots bottom-up, creating low-level behaviours, testing and locking them in, then adding higher-level behaviours until, eventually, human-level intelligence flowed seamlessly from the machine. The problem was that the second level induced errors in how level one functioned, and it took unanticipated debugging effort to get level one working with level two. By the time a roboticist reaches level four, the number of side effects overwhelms the original engineering effort completely, and funding dries up before success can be had. Once we begin bio-hacking, we are sure to discover side effects that the best simulators will fail to recognize unless they are equal in fidelity to the real-world. After how many major revisions will we discover that all our hacking time is spent trying to undo unintended consequences rather than optimizing desired new features? This is not a story of discontinuity, unfortunately, but the gradual build-up of messy, complicated baggage that gums up the works gradually and eventually becomes a singular centre of attention.

We may just discover that the Singularity, whether it gives rise to Full-Spectrum Superintelligence or to an Artificial Superintelligence, surfaces an entire stable of mediocre attempts long before something of real value is even conceivable. Just how many generations of mediocrity will we need to bridge and at what cost, to reach the discontinuity that is an existential matter of faith?

There is one easy answer here, at once richly appropriate and absurd. Pearce proposes that emotional self-control has one of the most profound consequences on our humanity, for we can make ourselves permanently happy. Learn to control the limbic system fully, and then bio-hackers can hack their way into enforced sensory happiness- indeed, even modalities of happiness that effervesce beyond anything our non drug-induced dreams can requisition today. Best of all, we could program ourselves for maximal happiness even if Me-2.0 is mediocre and buggy. Of course, this level of human chemical pathway control suggests a level of maturity that pharmaceutical companies dream about today, but if it is truly possible to obtain permanent and profound happiness all-around, then of course we lose both the condition and state of happiness. It becomes the drudgery that is a fact of life.

Finally, let us return to one significant commonality between the two hypotheses: they both demand that technology provide the ultimate modelling and simulation engine: I call it the Everything Engine. The Everything Engine is critical to AI because computers must reason, fully, about future implications of all state sets and actions. The Everything Engine is also at the heart of any IDE you would wish to use when hacking your genome: you need to model and generate evidence that your proposed personal modification yields a better you rather than a buggier you. But today, the Everything Engine is Unobtanium, and we know that incremental progress on computation speed will not produce it. We need a discontinuity in computational trends in order to arrive at the Everything Engine. Pearce is right when he states that the two meta-narratives of Singularity are not mutually exclusive. In fact, they are conjoined at the hip; for, if their faith in a future discontinuity proves false, then we might just need infinity of years to reach either Nirvana. And if the discontinuity arrives soon, then as Pearce points out, we will all be too busy inventing the future or evading the future to predict the future.