“The question whether machines can think as relevant as the question whether submarines can swim.” — Edsger W. Dijkstra
I love cooking, and eating, but for the carnivores among us its always depressing to hear how many years of our lives are being shaved off by our red-meat-eating habits. The herbivores are probably ecstatic every time the BBC news brings it up. Myself, being a guilt-ridden meat eater, like a smoker who can never quit, thinks we probably deserve all the bad karma. However, until guilt-free and life-enhancing test-tube-meat reaches a palatable stage, or our future generations are sucking mush through a straw into their chinless and toothless gobs, I will probably keep up my bad habits. Another guilt-inducing fact that is often pointed out to me (and why me, I don’t know; perhaps my guilt is written plainly on my sleeve for Buddhists to pick at) is how inefficient an energy source meat is; it takes so-and-so many fields to feed-so-and-so many cows to feed so very few lucky meat eaters. The ultimate limit of this logic of course is the fact that any bio-mass is extremely inefficient (both as an edible fuel source and as an renewable energy fuel source). To my mind the only solution to this inexorable logic is to cast-off our fleshly shackles and embrace a totally digital life powered by the most efficient fuel one can imagine. But what is the most efficient fuel? It is Max Tegmark.
— nom nom …nom?
Before explaining why Professor Tegmark is so efficient, lets discuss our brains a little. Our biological brains consume upto an amazing 20% of our energy intake. A substantial portion of this is spent opening and closing redundant ion channels on neurons, a kind of biological error correction needed because these channels are easily thermally or mechanically activated. It is thought that primates get some of their IQ advantage from having small, more closely packed neurons (evidence mostly consists of our lack of Cetecean overlords), but there is only so far this scaling can go; neurons need a minimal number of ion channels (for the error-correction redundancy) which ultimately limits their miniaturization. The ion channels could be made more reliable, so that there is a lower rate of thermal activation, but with an increased energy cost for those activations that we actually want. All of these are common problems for anyone working in information science, or anyone who has opened up a desktop computer. Specialized processing and parralization, via multi-core CPUs and GPUs, is becoming more and more necessary as we reach the limits of transistor miniaturization. But the communication overhead also imposes a limit on the usefulness of such compartmentalization. How do the two technologies compare? The human brain is estimated at 100 billion neurons, while the record CPU transistor count is 2.5 billion transistors, in Intel’s 10-core Xeon Westmere-EX.
— Can a 10-core Xeon processor enjoy a good steak?
Probably this is not a meaningful comparison, but it seems with human brains we have already reached the limit of a biological Moore’s law. So perhaps the only way forward, if we are going to improve cranially as a species (and be ultimately more fuel efficient) is a nice digital life. Imagining that we finally upload our brains to our Pentiums, or create the AI equivalent, how serious will our energy concerns be? Perhaps we will all be happy to live in virtual matrix-style World of Warcraft simulations. In this case our energy needs can be minimized drastically. First of all energy for actual processing can be almost completely eliminated. We can simply reduce the clock rate of our virtual noggins, assuming we no longer care about day to day life of the outside world, and of course our subjective experience will not be affected. Thus minimal solar power or similar alternatives seems sufficient. Still, what about physical degradation of our hardware and environment? It is a great irony that it seems we can never completely escape our physical incarnation. Similarly, if we do want to interact with our environment via robot bodies, real time, then our energy costs sky rocket.
— this is not a photo of Greg Egan. Also, wikipedia says he is a vegetarian!
When I was a teenager one of my favorite science fiction books was “Permutation City” by Greg Egan. In his novel virtual humans solve the problem of processing power and hardware by taking advantage of what he calls his “Dust theory”. It has been years since I read it, so I wont attempt to botch a retelling of his ideas, but a similar concept was proposed by Max Tegmark (http://space.mit.edu/home/tegmark/crazy.html). He posits that every mathematical system implies a physical universe, and if that system of sufficient complexity so that it can describe self-aware entities, then those entities will experience a physical reality. This concept comes with allot of metaphysical baggage about ensembles of mathematical universes, but it does suggest an interesting parallel to Greg Egans novel. Lets say we have a computer program that is of sufficient power to simulate a human mind (making the big assumption that that is possible at all of course), and we create a virtual environment with which that mind can interact, sense, feel, etc. If the program is entirely deterministic, based on a set of initial conditions, and relies on no input from our physical world, then from the subjective experience of the human mind within the simulation it makes no difference if the program is actually run or not. Every thought it can ever have is already defined by the initial conditions and the complexity of the program. So why run the program at all?
Lets avoid the issue of what that implies about our own self-determination in this apparently fleshy reality, as almost any sensible person would, and go back to our energy problem. Obviously the most efficient solution is to scan all our brains in such a way that they can be “run” as a computer program, define a sufficiently complex environment in which we can all virtually drink tea and discuss self-determinism in an ironic tone, and then never actually run the program. Of course at this point we (the apparently real ones) should all commit seppuku, as the main character in Greg Egans novel does. There we go, I told you Max Tegmark was the most efficient fuel. But to be honest I would rather eat a good steak.
One last thing. This article is mostly just for fun, but it does express one of my own personal opinions: that strong AI is possible, and that there is nothing so special about the machinery of the human mind. If you have a different opinion, please express in the comments!