mahiwaga

I’m reading I am a Strange Loop by Douglas Hofstadter right now, which is a sequel to his widely popular book Gödel, Escher, Bach. Hofstadter concerns himself with, among other things, the software of thought.

In the chapter he’s trying to make the distinction between hardware and software when it comes to thinking machines and thought processes.

This is important if you’re trying to posit that it should be feasible to implement intelligence on non-organic substrates such as silicon wafers. One can look at the human brain as a hyperconnected, super-complex, megacomputer, just like your Mac, except several orders of magnitude more complicated, with a design that is hundreds of millions of years old. But it is a distinction that has yet to be proven empirically. It may be that thought as we know it is dependent upon the hardware configuration upon which it is implemented, so you can’t just ignore the neurobiologists.

The reason why I lean towards the hardware-software interdependence is because hardware-software independence is a recent thing in computer design. I don’t think it was until the late ‘90’s-early ‘00’s that we had actual machines that could run multiple OSes with ease, whose parts were readily interchangeable. I still remember the 8-bit days of yore where you had to hand-optimize your code to cram functionality into 38k of RAM, and you had to rely on direct access to the hardware that the designers had hardwired into your machine. There were no abstraction layers. It was just you and the machine, with maybe 4k of convenience routines referred to as a kernel, and if you really wanted, another 4k for a language interpreter. But the higher level language did nothing to insulate you from direct hardware access. You still had to PEEK and POKE to get things done.

Still, you might think that Nature, having had several hundred millions years of a head-start on computer designers, would’ve figured out the abstraction thing and would’ve implemented cleaner designs with each iteration. Unfortunately, random chance doesn’t really work that way. What you usually end up with is a lowest-common-denominator design that random chance mucks around with every so often, and more often than not, that change will be fatal, but occasionally, it may be an improvement. It’s no coincidence that the basic plan of the human brain has been around for literal epochs, implemented in other creatures like dogs, velociraptors, frogs, and lampreys. And while absolute size has little to do with actual intelligence (after all, a baleen whale has a much bigger brain than a human), there are some morphological features that can be predictive of intelligence, like the surface area of the cerebral cortex, for example.

Hofstadter tries to argue that you don’t need to know the low level bits and pieces of the hardware in order to appreciate the software, and to a degree, this is a useful way to think. Unfortunately, his use of the heart as an analogy blows the argument to pieces.

If you ever want to piss-off a cardiologist, insist forcefully that the human heart is nothing but a pump. The pump analogy has indeed been extremely useful to physicians and physiologists alike, but we always have to remember that that’s all it is: an analogy. While we may model the pump function of the heart with an extremely simplified form of Poiseuille’s law (P=QR, pressure equals flow times resistance), it’s obvious that there are no rigid pipes involved, there’s a lot of non-laminar flow going on, mostly because the flow is actually pulsatile and not continuous.

Sure, the analogy helps physicians describe heart failure and pulmonary edema to their patients in simple terms, and the extremely simplified form of Poiseuille’s law allows us to study human physiology without having to resort to calculus, and to allow ICU physicians to make back-of-the-napkin calculations while they’re screwing around with pressors and inotropes in someone who is dying. But the pump analogy eventually fails, though. The most spectacular way I’ve seen it fail is in the description of heart failure.

The pump description describes systolic dysfunction as simply the heart’s inability to squeeze adequately to keep flow going forward. So the preload backs up, causing fluid build-up in the lungs, and in severe cases, the flow can’t adequately overcome the afterload either, causing ischemic symptoms like passing out. Given the pump analogy, for decades, physicians treated systolic dysfunction with cardiac glycosides, namely, digitalis. What digitalis does is essentially force the heart to pump harder. Physicians also were taught to avoid medications that would weaken pump function, medications such as beta-blockers.

Well, wouldn’t you know it, when they studied this more closely, it turns out that while digitalis can improve symptoms in patients with heart failure, thereby decreasing the number of hospitalizations, it did nothing to change mortality rates. In stark contrast, it was found that the addition of a beta-blocker not only decreased morbidity, it also prolonged survival time. How counter-intuitive is that?

The reason why this probably works is because of the molecular physiology involved. You cannot avoid having to study signal transduction pathways if you want to understand how to treat heart failure. It has been discovered that the neurohormones epinephrine, norepinephrine, angiotensin, and aldosterone can bind to receptors in the heart and activate the TGFβ pathway. What has been recently appreciated is that the reason why pump function continues to deteriorate in heart failure is because of remodeling. The heart becomes more fibrous, less muscular and elastic. How is this mediated? By growth factors like TGFβ.

It is probably not coincidence that we have since discovered that beta-blockers, angiotensin-converting-enzyme inhibitors, angiotensin receptor blockers, and aldosterone antagonists all help with not only improving the quality of life of heart failure patients, but also with increasing their life expectancies. It used to be that a diagnosis of heart failure had a similar prognosis to metastatic cancer, but now people can live for years and decades. And we would never have figured any of this out if we had stuck to the pump analogy of heart function.