by Michael Brunnbauer, 2012-09-21
Raymond Kurzweil thinks there is a very good chance that he will live forever. Accelerating change in technology will prolong his life up to the point where mind uploading becomes possible, the famous inventor says. He predicts this point to be reached between 2030 and 2040 and has developed a rigorous program to stay healthy until then - which includes taking 200 pills per day. He is selling books promoting his ideas and his health program. And he is not alone. A lot of people - especially geeks - are dreaming this dream that converges in the broader Transhumanism movement.
Raymond Kurzweil is very smart. Smart enough to know that "forever" really means: Until the heat death of the universe - notwithstanding unexpected "accidents". Fair enough. The prospect of living, say, 10^40 years, is quite alluring to many people - including me.
But whenever prophets show up and start talking about a rapture in the future and how you should change your life to be prepared for it, you ought to be sceptical. And this is how more sceptical people call the idea of mind uploading: The Rapture of the Nerds. I will try to explain here why I think that nobody of us will live to see mind uploading and why it does not make sense.
While the idea that the human mind can be seen as a deterministic machine is still offending to many people - including me some years ago - most of us got used to it and I will not debate this point. Some are trying to save non-determinism in the brain with quantum theory and they may even be right but the advocates of mind uploading will argue that we may use "quantum hardware" to simulate this aspect.
I will also not argue about desirability of living for extremely long life spans but let me say that it would not surprise me if it turns out to be more like hell instead of heaven.
The feasibility and time scale of mind uploading or "whole brain emulation" has been debated scientifically in depth. Have a look at this report
The term "whole brain emulation" is misleading. Simulating a human brain without body and environment will get you a mad brain within a very short time range. There is a project at openworm.org trying to simulate the nematode C. Elegans whose body and neuronal circuits have been mapped out in great detail - every one of the ca. 1000 cells of this creature has its own designation. Estimates when this project will succeed range from 2 to 15 years and others have tried for at least 15 years! http://www.artificialbrains.com/openworm says that "The mechanical model of the worm requires about 5 teraflops of computing power. The muscle/neuronal conductance model requires about 0.24 teraflops".
The brain is in a constant feedback loop with its environment and in order to see if the simulation of the worm brain is correct, you need to simulate the body and the environment too - and the simulation of the body/environment consumes much more computing power than the simulation of the brain!
I don't know if human minds will get mad in a simulated world with the detail of World of Warcraft but there is at least the matter of aesthetics. And this matter does not really go away with artificial bodies that roam in the real world.
The ability to start, stop, accelerate, decelerate, clone or modify simulated brains poses serious ethical questions that are probably not really relevant at this point. What is relevant is that the chemistry of the brain is involved in its computation.
Assuming that something like the Hodgkin-Huxley model of neuron chemistry is detailed enough for a human brain simulation (a bold assumption), you get a set of differential equations that can only be solved by iterating over discrete time steps. Simulating a single neuron with its complex network of dendrites, axon and synapses requires breaking it into many compartments each represented by a set of differential equations with different parameters. Such a simulation of one neuron needs all the computing power of a modern laptop and we need to simulate 80-120 billion neurons.
What I am driving at is that as long as we do not understand the higher level functions of the brain fully, its simulation requires resources that are many orders of magnitude above those of the real thing. For example, the human brain consumes only 20W of energy. This gigantic overhead of simulation will probably not go away as long as we emulate chemistry with something else.
The same argument about emulating nature on top of nature also probably holds for an aesthetic simulation of body and environment.
This means that if we prolong one life by mind uploading, we use up resources that could be used to sustain the lifes of many real humans. Even if the uploaded brain does not use more resources than the real brain (which I think is impossible), the resources it uses during its extremely long life span could sustain many "old-fashioned" lifes. Do we have the right to prolong our life at the cost of unborn life? To some degree, yes. But the degree involved here is mind-boggling.
So as long as we do not dismantle other planets to build supercomputers, mind uploading is unethical. And if we can dismantle other planets, maybe we can also terraform them or make some other use of them for real humans?
What if some day we understand the higher levels of brain function and can replace chemistry emulation with something simpler? I am certain that the human mind cannot fully understand itself. We could try simplifications by trial and error but how do we know that some essentials don't get lost? We are deep into the Qualia problem here but to be honest: This problem is there on every level of simulation. It just gets more "pressing" the higher the level of simulation is.
Another aspect is change. Learning, change and evolution (via birth and death) are important aspects of life. Memory and learning in the brain requires constant rearrangement of the synaptic network. The article about dendritic spines in Wikipedia says that 10-20% of them spontaneously appear or disappear on the pyramidal cells of the cerebral cortex in a matter of hours. We are far from understanding these processes and having a model to simulate them. But if we can simulate that one day, will it be enough to prevent a dead end?
Will we live long enough to see mind uploading? I don't think so. Even if our life span will be prolonged to 120 years or more by technical progress, some very difficult problems will have to be solved in that time:
Raymond Kurzweil tends to point to exponential functions of technical progress to dismiss such arguments but he does not mention that technical progress is a series of exponential functions that break down at some point and are replaced by new ones when the next big innovation occurs. Moore's law *will* break down at some point and we don't know if or when there will be some break through after that - so that exponential growth continues.
Biologist P. Z. Myers has written some excellent blog posts about Kurzweils theories here, here and here