Sunday, August 16, 2009

Kurzweil's Horcrux

Ray Kurzweil is a noted innovator and futurist, who has been consistently making the right technological predictions for a couple of decades. Recently, he has started making certain predictions that many people considered outrageous and way beyond the mark.

His predictions are about what is called the technological singularity, the point in time where machines become so intelligent that they don't need any human intervention at all for the further advancement of technology. Kurzweil predicts that this will happen in a matter of a few decades. The reasons for his optimism have nothing to do with our scientific progress in cracking the hard questions of complexity theory or artificial intelligence, but with the explosive growth of processing power in computer chips.

He has written several books and essays on the internet, highlighting this point. But I'd like to quote his argument in a few lines below.

" By 2029, sufficient computation to simulate the entire human brain, which I estimate at about 1016 (10 million billion) calculations per second (cps), will cost about a dollar. By the mid-2040s, the non-biological portion of the intelligence of our human-machine civilization will be about a billion times greater than the biological portion (we have about 1026 cps among all human brains today; nonbiological intelligence in 2045 will provide about 1035 cps). "

Kurzweil believes that when it is possible to computationally simulate an entire human brain, it will also be possible to download and save the configuration of a brain onto such a computer. Brain imagery and sensing have been making some progress in recent years, and it might be even possible to download an entire human brain onto computer memory, making Kurzweil's prediction come true. Kurzweil believes that this will solve the hard problems of AI, because the computer then would be capable of solving any problem that a human brain could solve.

For want of a better word, I would like to term this computational copy of a brain as a horcrux - an example of dark magic mentioned in the Harry Potter books. I believe that this concept is a very instructional tool in musing about consciousness, immortality and other such philosophical issues. So I'd like to exploit this connection in today's post.

The question I would like to pose is "Would a person achieve immortality when his neurological state is copied into a horcrux ? "

Kurzweil answers this in the affirmative. But I think he's wrong about this. To explain this further, let me introduce another philosophical thought experiment.

Imagine that aliens attacked earth and they cut off your hand from your body and kept in a glass-jar. If you look at this hand, would you say that it is "you" or just "your hand" ? Most people would reply it's just a hand, and not "you". Imagine that the aliens got more aggressive and cut off your brain from your face, wired cameras to your brain and oriented the cameras towards your face. Would you say what you are looking at is "your face" or "you" ? This is a tougher case, but when you think deeply enough, you would say it is just "your face" and not "you.

The reason for your reply is that even if somebody destroys your face, "you" would still live. "You" would keep on getting your sensory input and you'd be processing your thoughts. So your face is not "you".

Now let's ditch the aliens and get back to the horcrux. Imagine that you got your brain copied into a super-computer. Where would be "you" ? Would "you" be sitting in the computer or in the human body of yours ? Or would you be sitting in both places at the same time ?

I say that "you" will be sitting in your human body. If your body is destroyed, "you" would die. If the super-computer is destroyed, it wouldn't harm "you". (A further clarification of my position comes later). Kurzweil thinks otherwise. He thinks that "you" would be sitting "both" in your human body and in the computer. Even if one of them is destroyed, the other will survive. So, "you" would still live, achieving immortality in this fashion.

Kurzweil says that, irrespective of our theories, it will be possible to conduct such an experiment at some point in the near future, and thus we'll get to know an answer backed up by experiments. This will be the first time in human civilization, that a scientific experiment is conducted to deal with the questions of consciousness. Until now, consciousness has been a subject that's investigated only "internally" by meditation conducted by an "internal" eye. Soon, it will be possible to conduct such experiments on an "external" scale.

This is fascinating, and I think that such experiments will lead to some deep breakthroughs in physics and biology, apart from computer science.

Personally, I subscribe to the ancient Indian theory of Advaita, which says that consciousness is singular and universal. It says that it has no birth and no death. My individual consciousness in this human body is like a wave in the ocean of the universal consciousness. My human body is just like a particle that is being moved by this wave. What is moving "the wave" is the ocean, which is eternal and omnipresent. It is this wave that creates the sensation of "I" in my human body. This is consciousness.

Advaita says that what is speaking inside me and what is listening inside you is one and the same thing.

So actually, my position with the horcrux stands as that "you" would be sitting neither in your human body nor in the horcrux. "You" are omnipresent, universal and eternal. But this "you" is fooled into thinking that it is just confined to your human body. So after your brain is copied into the horcrux, this consciousness & its associated illusion would be still sitting inside your human body. Your horcrux, if it could indeed be created as an intelligent agent through copying your brain, will be an independent agent, powered by the same universal consciousness but suffering from a similar illusion like how "you" in your human body would be suffering.

According to the theory of Advaita, consciousness cannot be copied. It exists only in the singular. A good analogy is with language : a "word" can be copied, but its "meaning" cannot be copied. A "meaning" can be copied only by copying the person who understands this meaning. Thus, a "meaning" doesn't exist beyond the very consciousness of a person.

There is a certain amount of scientific backing for such a theory of consciousness. This comes from Heisenberg's uncertainty principle, which says that for a quantum-particle (like an electron or a photon), only one of the variables of position and momentum (velocity) can be known. The very instant one property is measured, the other becomes indeterminate. Certain neuroscientists and physicists, such as Dr. Roger Penrose, believe that consciousness is a quantum mechanical phenomenon inside the human brain which operates through the so-called microtubules. If this is the case, then Kurzweil's dream might never come true. No cerebral photography can ever be able to copy the entire state of the human brain because such a thing is ruled out by the theory of quantum physics.

However, it might be still possible to build a horcrux, which will be a self-conscious entity which is capable of intelligent behavior. But that will just be a copy of "you" and not "you". It will be a copy of all your emotional and mental weaknesses.

Apart from the uncertainty principle, there is another phenomenon of quantum physics that is equally baffling, which is known as quantum entanglement. Several times, a pair of particles is entangled in the quantum world, at the very moment the particles are created. If you measure the quantum property of one particle (say the spin of an electron), the property of the other particle is automatically determined (the spin of its paired electron). This can be checked by another experimenter located several miles away at a much later point in time. This entanglement principle has been validated several times through experiment, and this is explained by various interpretations of what's happening, all of which are quite "weird" and counter-intuitive. Some of these interpretations (based on hidden variables) are less likely than the others, though they are not yet totally ruled due to statistical limitations in current measurements.

The entanglement principle says that either (a) reality is an illusion or (b) free-will is an illusion. Many physicists settle for the option (b), though it looks totally counter-intuitive from the perspective of a human being who can definitely sense a feeling of self-consciousness.

What Advaita says is that consciousness is real, but it is universal. So it is only one single person who is measuring the property of the first electron, and later that of the second electron. It is not two different experimenters even though the measurement is made from two different bodies. This way, Advaita dissolves the paradox by tying free-will to reality.

According to Advaita, free-will can only be defined upto a scale. An entity (whether that be an electron or a human being) is "free" only up to the scale of objects that it is "aware" of. Freedom comes from awareness. There is no meaning for one's freedom beyond one's knowledge.

Similarly, according to Advaita, reality doesn't exist beyond mutual agreement of the two particles. In other words, you are as real as I am. And both of us are real because we see each other. Take away this mutual awareness, and we shall cease to exist.

I like the theory of Advaita (and its parallel philosophical systems in Hindu religions) very much. But until now, they have only remained at the level of hearsay and speculation. But soon, it will be possible to conduct experiments and test the various theories behind this mysterious phenomenon called consciousness. Isn't it fascinating ?


Ian Parker said...

Not all Kurtzweil's utterances have proved true. Kurtzweil talked about nobody knowing a word of Arabic yet Google Translate was the best translator out.

Looking at some of the translations all claims for AI disappear. Google translate does not even know enough to get equations right, and frequently screws up.

- Ian Parker

Ray Lightning said...

Hi Ian,

Machine translation across different human languages is a very tough problem. I've worked in it for a few years, so I know how bad the problem is. This is one of the so-called *hard* problems of AI, and I don't think it'll be solved anytime soon.

The crux of Kurzweil's (new) argument is that instead of coding artificial intelligence into a machine, if we can copy a human brain into the machine, then the machine wouldd be capable of doing all the things that can possibly be done by humans. This argument circumvents the hard problem of AI, but rests on an (untested) assumption that copying a consciousness from brain to a computer is indeed possible.

Nicholas Sagan said...

The brain holds consciousness, so your face and hand analogies do not apply.

Ray Lightning said...

What's the proof Nicholas ? Neuroscience is still *far* away from understanding what consciousness is, forget about pointing its seat of location.

In my opinion, the question of consciousness is related to the so-called "P=NP ?" problem in computer science. We humans, even animals, perform such complex tasks which are known to be exponentially hard for a computer (deterministic Turing machine) to solve. One such problem, as mentioned by Ian Parker, is translation across languages. There are various other problems as well. Having an understanding of these so-called hard problems of AI helps in understanding what consciousness is, and what are its artifacts.

There are several physicists (Roger Penrose, who I mentioned in my post) who argue for a quantum mechanical origin of consciousness. Neuroscience is still far far away from understanding this.

Nicholas Sagan said...

Neuroscience is past the point of understanding how consciousness emerges from neural networks. See Jeff Hawkins work for some structures involved.

Machines are already smarter than some humans. That rate is only increasing.

Penrose and his 'What the Bleep do We Know' comrades' aren't publishing any decent papers in the field.

Ray Lightning said...

Machines are not smarter than humans.

The reason is that a linear increase in processing power cannot cope with the exponential nature of complexity of certain problems (known as NP-complete problems). These are the so called hard problems of AI. Just throwing in more processing power wouldn't do the trick if we use the very same algorithms.

Neural networks are used to explain pattern recognition, not consciousness. I recommend you study the nature of consciousness and its artifacts. For example, no computer has yet managed to achieve human-level translation across languages, even though it has far more processing power and memory than is available at the disposal of a human being. This is also the reason why no computer has yet won the Turing Test.

Do you work in neuroscience ? Otherwise, please resist from making claims such as "neuroscience is past the point of understanding how consciousness emerges". Because such claims are actually outrageous and cause quite a commotion in people working in this field, or in cognitive science, psychology or computer science, who study these questions in detail.

Nicholas Sagan said...

Terri Schiavo was dumber than most machines. Many mentally deficit people are dumber than the best AI.

We sure have come very close to winning the Turing test, and have convinced some participants that the computer is a human.

I'm a masters student in computer and aerospace engineering; and employ self-directed intelligence in avoiding terrain and objects within these UAVs.

To repeat, your hands and face arguments are fallacious, as they have nothing to do with copying consciousness, which is entirely materialistic and located in the skull.

Ray Lightning said...

Actually, I used the analogy of hand and face removal by aliens to just give an idea of the nature of the problem. To understand what consciousness means. This analogy helped me define consciousness in the next paragraph, where I explained it through the phenomenon of death.

I don't want to generalize the argument from hands and face to the brain. I just posed the question.. whether consciousness be copied by copying the neural state ? Your answer says yes. My hypothesis is a no. But we should both accept that this is an open problem, we do not know for sure what will happen.

I wish you luck with your algorithms on obstacle avoidance and navigation. This is a very hard problem, especially if you want to do also path planning.

Now compare this with a bird that flies in space, navigates through a forest and lands perfectly on a rock. The size of the bird's brain is not enormous, but it does this navigation without getting stuck into the problems of exponential complexity that computer programs will get stuck in, if you use the algorithms that exist today.

In fact, I am a research student working in computer graphics and computer vision. I know a bit about the nature of the hard problems of AI. From the perspective of a computer scientist, I can say with confidence that we are nowhere close to solving these problems than we were 50 years ago when it all began.

FallowCypher said...

Gentlemen, your 'sciencentrism' shows. Consciousness has been proven by as many cultures as there has been to prove it throughout history. Speaking as if it does not exist or it cannot be explained is just that: speaking of it - a limitation of language. Is the paradigm of paradigms another paradigm? If you consider consciousness as the source of all that can be employed to understand it, then the problem disappears. So, the face and hand analogy is valid when employed by one, and not when employed by another whose consciousness cannot conceive of it employed as such. The contradiction is the illusion.

Incidentally, I'm a Taoist and have been in computer science close to 40 years now.

監控 said...