Wednesday, September 1, 2010

Reverse engineering the Brain (insert Pinky and the Brain reference here)

Found this via CETMONS Facebook page here:

Color me confused. Reading this article closely..

I can’t quite figure out what Ray Kurzweil is on about. What exactly does he think reverse engineering is, and what does he think is required to carry it out? Finally, what exactly is the relationship that exists between computer based simulations and reproduction of brain functionality?

The folks at CETMONS helpfully included a link to someone who thinks Kurzweil is (a) a kook, and (b) confused.

So, that makes me feel a bit better. I’m just as confused, it seems.

Kurzweil is someone who thinks something called the “singularity” is just around the corner. No, this does not mean that we are about to be sucked into oblivion or some parallel universe by a star-roving black hole, but he does think we are on the verge of creating AI that is smarter than us, in some global sense of the word “smarter,” and that that AI may ‘take off’ on its own, and yes TAKE OVER THE WORLD (insert Pinky and the Brain reference here) or enter into some symbiosis with us, initiating either a utopia, or a dystopia.

Problem is, I can’t quite figure out what K’s basis is for this prediction, and what exactly it is that will be doing the world domination (or kumbaya symbiosis). Let me see if I can spell out my confusion by addressing the three questions I opened with.

1 & 2: What is reverse engineering, and what does K think it is?

As near as I can tell from my always reliable wiki sources,,sid183_gci507015,00.html

reverse engineering involves carefully taking apart and/or otherwise analyzing an artifact or other functional object (such as an eye) in order to understand how it functions as a system or organized whole. This will also enable you to (at least in theory) create similar parts and build one of the gizmos yourself. This may or may not land you in legal hot water. It also allows you to trial run modifications of the system, via modifications of parts, and relations between parts.

Once you have a thorough understanding of the works of the thing, you can tinker without tinkering in the dark. You have an idea what your modifications will do to the overall function.
One can also reverse engineer software, which involves diving into the program code (the program language) in order to get at a more basic level, the source code. Now, I’m not going to pretend I know what the difference is here, but, once again, it becomes possible to replicate the source code, the upper level superstructure program that contains it, and perhaps modify both.

One can use computer simulations to reverse engineer from object to blueprints. For instance, if I want to have a detailed schematic of some gizmo, say a competitor’s car, I can buy one, take it apart, use some measuring devices to input dimensions of the various parts, input values that mimic physical properties of the parts, properties I can investigate and discover via traditional lab analytical methods, (properties such as tensile strength, etc..) and can eventually end up with a detailed blueprint of the whole car, along with a set of working hypotheses concerning what constraints guided the original engineers to create the parts using the particular materials and dimensions they used. I can do all this even if the original blueprints are long gone or jealously hidden. With enough such information, we can even do some detective work in figuring out what purposes mysterious contrivances have (consider the Antikythera mechanism. Quite a lot of reverse engineering has gone into figuring out its blueprint and its purpose).

That, in short, is what I gather to be ‘reverse engineering’. Is it what Kurzweil is talking about? I don’t think so. He seems to think that running a computer simulation of the brain is in all practical effect building a brain. Let me explain, noting an ambiguity pointed out by Mr. Myers, kook slayer.

According to the first half of the WIRED article, effective computer simulation of the brain will involve thoroughly mapping the myriads of connections between neurons. Yep. Seems to fit the above description of reverse engineering. What is more, that simulation should also make efforts to build in the physical properties of (or rather simulations thereof) the neurons. This will all have to be emulated in computer code, just as tensile strength of a rear axle made of some particular kind of metal, must be represented by automotive engineers using CAD as they go through the process of designing that part. So, as far as it goes, this doesn’t seem that controversial.

But, when the article gets around to discussing Kurzweil things get a little weird. He says that the information needed to simulate the brain need not to be extracted from examination of the brain, but instead is to be extracted from a careful analysis of the human genome. His reasoning seems to be that since our genetic material is something like the ‘machine code’ for construction of human brains, then all the info you need about those brains is already ‘in there’ so to speak.

Myer rightly points out that all the information you need to simulate a brain is not in that code. While that code does provide direction, it will not provide detailed neural maps, for the simple reason that actual neural circuitry is the result not only of DNA code, but complex interactions of DNA with cells, neurons with neurons, nervous system with other systems within organisms, and interactions of all of these with environmental factors. There is a complex biochemistry of the brain that you would be hard pressed to read off the genetic code in isolation.

In short, problem #1 with Kurzweil: the first half of the article seems closer to the mark as presenting an accurate picture of what reverse engineering a human brain entails. Kurzweil misses the mark by comparison. This brings me to question #2.

3: What is the relationship that is supposed to exist between computer simulations of neural activity and reproduction, (or should I perhaps say, literal creation) of something that has full brain functionality, that is; something that can do all we brain possessing beings can do, and perhaps more?

From what I can gather, some folks think that a thorough computer simulation of a functioning brain (one that represents or simulates in computer language or code not only the structure of the human brain, but all the various physical and biological parameters) will ‘sprout’ full blown functionality, just as a real brain apparently sprouts mind as an emergent property of its physical behavior. But, why should one believe that this would happen? Not to sound to snarky here, but no automotive engineer I am acquainted with believes that it follows causally, or nomologically, from the fact that one can exhaustively emulate a Corvette that one thereby owns a Corvette. If only that were true! But, as near as I can tell, it seems that folks such as Kurzweil are beguiled by just that metaphysical leap of faith. I simply doubt it. Even if it were possible to thoroughly emulate Kurzweil in a computer program (sim city with a vengeance), all of him, not only his brain, but his entire body, DNA included, it does not follow that we have created a second Kurzweil, or even a second person.

And because I just cannot help myself..