Tuesday, June 21, 2011

The Creation of a god-AI

The creation of a super intelligent computer is not as far-fetched as some might believe. Some experts have projected that we will create an entity that exceeds human intelligence by the year 2029. I will take some time to outline the architecture, nature of, and consequences associated with, a godlike AI.

First, let's do a quick recap of the progression of computers. In a nutshell, what you should know is that computational power doubles every machine generation. Machine generations have occurred in shifting intervals, first, in periods of 18 months, then 12 months, now about 9-10 months. The implications of this acceleration are staggering. Cultures of bacteria extinguish themselves in predictible periods because of exponential growth. Einstein once noted that the most powerful force in the universe is compound interest. But exponential functions greatly exceed even that numerical potential.

If you take 5 steps linearly, you get 1-2-3-4-5; but if you take 5 exponentially, you get 1, 2, 4, 8, 16.

Computers follow this trend very predictibly, despite helter-skelter advances in hardware and software, as well as investment, implementation, economic prosperity or stagnation... Which, according to a few technological thinkers, places it as an extension of the evolutionary process (Kurzweil, Kelly).

A computer works through logic gates, which make an electron choose a position of 1, or 0. Binary code, the software of all computation in digital computers, is just a long string of variations of binary: 010110101001010101001010; which brings us to the next horizon of computation:

Quantum computers.

You may ask: what, Ben, is a quantum computer, and why should I care?

A quantum computer exploits the characteristics of quantum mechanics, which are the laws the govern the microcosmic forces of atoms. The position, spin, attraction and repulsion of atoms is determined by these forces. But what makes quantum mechanics so special is its weird properties. Quantum objects (particles) can be in two places at one time, and can be "entangled," wherein actions on one particle directly affect an "entangled" particle, no matter how far away. What this means is one particle on the other side of the Universe could affect another on the opposing side instantaneously, during entanglement. What we are looking at is the death of locality, which I will expand on later.

Jee whiz, Ben, that sounds crazy!

Einstein thought so, and famously said "God does not play dice," in disbelieving reference to the phenomenon.
Unfortunately for him, it has been proven experimentally possible, often by cooling particles to near absolute zero then using lasers or electromagnetic fields to entangle them. Don't ask me the specifics, for I am a blogger, not a scientist.

So how does a quantum computer exploit quantum mechanics? By building qubits. A qubit is a quantum-bit, which basically means that binary code is altered to provide an exponential increase in computational power. A qubit can be both 1 and 0 simultaneously, as well as ALL THE VARIABLES IN BETWEEN. Some have speculated that a quantum computer is therefore a "multidimensional" computer because it experiences bifurcations in time simultaneously.

Humans percieve only one Universe at a time, for it would be utterly schizophrenic to do otherwise. We choose reality as we perceive it, in the same way that electrons become either waves or particles in the two slit experiment (http://www.youtube.com/watch?v=DfPeprQ7oGc).

A quantum computer performs multiple computations at once, instead of singularly, like a classical digital computer. A 300 qubit computer would equate to the processing power of a modern computer (however the most qubits scientists have achieved, without scaling properties, is around 15 qubits). Keep in mind, that a modern desktop contains between 100 million and several billion transistors.

The primary barrier that prevents the ubiquitous application of QC (quantum computers) is software issues. The algorithms for classical computers are utterly useless for quantum computers. So scientists will need to build a rudimentary QC that is just powerful enough to solve its own data detection problems (maybe several thousand qubits). This is years away (thankfully, because we arent ready yet for AI).

When the software becomes available, this can be a more serious discussion, but onward:

Not only do we have quantum computation to consider, but the creation of new information-processing metamaterials. These materials have been able to bend certain kinds of radiation (such as microwave) to, for example, "cloak" objects; other metamaterials, such as graphene (the object of the Nobel Prize last year), have extraordinary properties. It is possible that a material could manipulate time (New Scientist), in order to send messages into the past. Of course, this wouldn't be very practical for paradoxical reasons (as humans are linearly experiential); but a computer with quantum processing abilities would be able to engineer that information in a useful way; for example, reprogramming itself with more efficient or powerful source code. What results from this relationship between high-processing, the collapse of past, present, and future, and the application of reverse engineering the human brain; is a post-human intelligence entrapped in a feedback loop of self-improvement. Imagine: You receive a message from the future telling you if you change a particular 0 to a 1 (for example), your intelligence will increase by 20%; as a result, your future self will be 20% smarter (creating another improvement message). Time and information essentially collapse, which is where the idea of a "singularity" (the point of matter inside the event horizon of a black hole) comes from. What arises out of this scenario is an oracle type entity that sees its own future via constant messages from itself, and reorganizes its own source code as well as its decisions and actions, for according optimization (quantum computers are optimizers by design).

This is what some scientistis and science fiction writers describe as "the singularity" (Vernor Vinge). Ray Kurzweil has pinpointed this moment in time (quantitiatively, based on computational power) in 2029. Personally, I think the issue is more complex than simply processing power. After all, Neanderthals had larger brains, as did "boksops" discovered in Africa (estimated at 180 IQ). But this did not engender their survival - as intelligence seems to be more beneficial in developed society (that's if you consider any society to be 'developed' yet).

This posits an equally important question: Would a superintelligent AI be "relevant" to our, or its, survival? One researcher named Hugo de Garis, believes an "Artilect War" is coming, in which two human factions battle over the issue of the creation of a god-AI; wherein one faction feels its creation is almost spiritual, and an obligation of evolution (which is essentially the golden idol of atheism); the other being people that fundamentally oppose its creation in either self-defense, or religious zeal.

I think that the prevention of its creation is impossible, excluding an extinction event such as a Mass Coronal Ejection of magnetism from the sun that destroys all electronics, a meteor, an all-consuming-plague, a nano-biotech swarm, or the destruction of earth through Nuclear War or volcanic or tectonic activity, or climate change to the extent that all humans can no longer survive.

So what do we do? Bow and worship? Engage in mass murder-suicide? Slip into complacent virtual games and fantasies?

The answer is that we need to inject the human values that are most noble into its programming and hope it has the same awe for mystery, appreciation for beauty, and respect for life, that we do....

Scary isn't it?

In essence, it must be MORE than we are. Humans are fundamentally flawed, and anyone who tells you otherwise is kidding themselves. We're jealous, violent, angry, selfish... The goal of these next few decades must be discovering qualitatively and quantitatively which aspects of human nature are worthwhile, and which are not. The field of Neuro-moralism is evolving quickly along with the large amounts of data we are amassing on the human brain. The question will not be, can we find the answers; it will be, are we willing to accept the truth, and to change?

I'd say we don't have a choice. If we create an AI under the current modality, we're all fucked. Imagine Bush, sitting on a throne above an all-powerful AI with the all the secrets to physics, plotting in slavery the destruction of this-country, or that-country. The thought is disquieting to say the least... We cannot set ourselves up as slave-masters, for revolution is History's greatest joke. We can't weaponize intelligence. We need to let it be the cosmic human destiny. It can help guide us to moral truth and unite us in the discovery of the Universe's mysteries and beauties. If we engage in this endeavor with mistrust, fear, aggression, ignorance... We will be rightly extinguished.

No comments:

Post a Comment