Saturday, July 10, 2010

Kant and the cognitive bootstrapping problem

It turns out that our brains come pre-patterned to deal with space, time, causality, language, and morality, among other a priori concepts. Where did they get them? From evolution, of course.

Respect for dead philosophers is an odd thing. They are often revered for introducing ambitious systems of esoteric metaphysics. Occasionally for having novel ideas. Sometimes they even write well. The questions they raise are perennial and to some, fascinating. But rarely have they solved anything. Ambitious systems are propounded in one era only to be forgotten or overturned in the next, at best remembered by a catchphrase, like "God is dead", or "thesis, antithesis, synthesis". It may be a brutal disservice to the history of ideas to encapsulate the multi-volume cogitations of past philosophers in such token fashion. But unless they have supplied solutions to some of the perplexities of existence, what else can we in all honesty do?

Immanuel Kant grappled with our capacity to reason, among much else. He realized (as prodded by Hume) that experience alone was not enough, combined with reason, to organize our thoughts about the world. We need to have some pre-existing template into which to pour all those experiences, like a basic grid of space, time, and a sense of causality, if we are to have any hope of getting off the ground. Which is to say- our minds can not be an empty slates if we want to learn anything. Or to put it another way, we can not epistemically pull ourselves up by our own bootstraps. Thus he came up with a priori concepts, which he took to be given structures of the mind, however the mind comes to be.

From the wiki pages: "... Kant is thought to argue that our representation of space and time as a priori intuitions entails that space and time are transcendentally ideal." ... "According to Kant, a priori knowledge is transcendental, or based on the form of all possible experience, while a posteriori knowledge is empirical, based on the content of experience." ... "Something is transcendental if it plays a role in the way in which the mind "constitutes" objects and makes it possible for us to experience them as objects in the first place."

Despite Kant's other ventures in defense of god and of free will, this aspect of his ideas did not neccessitate any opinion on how the structure of our minds arises, and is thus quite consistent with modern cognitive psychology and developmental psychology.

Two recent papers deal with development of spatial consciousness in rats. Incidentally, another recent paper shows that this system (spatial representation and memory across the entorhinal cortex and hippocampus) which is so well studied in rodents, works the same way in humans, studied by way of strokes that create transient dead spots in the hippocampus. The rat papers study very young rats, just as their eyes have opened and they have their first exploratory experience outside the nest, to tell whether their hippocampal and entorhinal place memory/sensing systems are already in place, or only form later after further experience. Their conclusions agree that, while experience helps this system grow both physiologically and in behavioral accuracy, it is in place from the earliest times.

The hippocampus is the location for storing temporary memories, and also for place memory. It is telling that a popular memorization trick among the ancients was to vizualize a palace with endless rooms, where arbitrary memories are stowed by location, available for retrieval later. In the contemporary world, graphical metaphors lead the way in computer user interfaces, giving us "windows" on virtual spaces that organize our affairs far more richly and naturally than text indexes, file trees, or other more easily-implemented methods.

The hippocampus gets its location input from the entorhinal cortex, which is in turn connected to the inner ear, vision, and other sensory systems. The entorhinal cortex contains neurons that respond to head orientation, spatial grid at various scales, borders, and absolute location. This means that scientists have been able to record the activities of neurons that fire only when the rat faces a particular direction relative to its environment, or crosses an imaginary line in an arbitrary polygonal grid that divides up its environment, nears a spatial border, or hits a particular location or landmark.

These cells are assumed to be directly involved in the animal's consciousness of place, since their removal dramatically impairs their ability to learn mazes and know where they are. The question then is... when is all this information set up? We know that learning creates new grids, landmarks, and borders all the time. But is there anything in this system at the start, as neurons get wired up? These papers claim that yes, there is.

They inserted electrodes into the relevant location of 16 day-old rat pups, and found that already at this time, two days after their eyes have opened, and before significant mobility, they have head direction cells and place cells, and one paper finds grid cells as well at this early time.

Example from Langston, et al. showing the location of probes in the entorhinal cortex (A), and local neural firing rates (with peak firing frequency noted) of presumptive head-direction cell recordings, plotted vs actual head direction (B).

Unfortunately, for all the technical pain of doing this kind of work on such young animals (for all concerned!), I don't think they did this early enough. Even with their eyes closed, infant rats are probably acutely aware of place, if only to find the mother's nipple. Vision is not the only input to the entorhinal cortex, so doing this work soon after the eyes open seems insufficient to say that the overall space orientation system is in some part natively present by genetically-driven development alone. Indeed, even in the womb, babies probably have some notion of space, since they are kicking around and active.

As Langston et al. conclude, "Whether the formation of prototypical representations in the parahippocampal and hippocampal cortices requires translational or vestibular experience at younger ages, in the next, remains to be determined."

So the ideal experiment would seem virtually impossible to do- to isolate a developing rat's entorhinal cortex (and thus hippocampus) from external influence to see what the brain wires up all by itself. This could be done by severing the inputs in such a way that they could not regrow. Which would probably derange its development substantially, by cutting off the nerves invading from the sensory areas. Would that tell us that modest inputs from very early times may have strong influences on the nascent wiring? The hard part would be to measure the naive neuronal maps- if they have no input, then how can we measure them? How can a grid cell fire if the rat can't detect a grid? At that point, we might have to resort to dreams as a way of probing the native capabilities of the brain. Could a rat with no spatial experience or sensation still have spatial dreams? Now that would be an interesting question!

In the end, this work seems at best indicative of the conclusions they claim. I would generally agree that the Kantian a priori concepts of space and much else are pre-wired in the brain in some way, based in turn on the evolutionary patrimony of embodied knowledge gained through the school of hard knocks. But the very concept of brain "wiring" is transgressive, both carrying information and constituting information derived from past development and genetic influences. Mammalian fetuses experience things from some early point in time, based to various degrees on genetic wiring. From there on, development is a mad dash of experience and genetic patterning each feeding off the other to generate that mighty device- the fully formed brain.

I realize that this work does not even try to address the a priori concepts of morality, time (also here), social relations, language, etc. I believe they follow the same pattern of modest pre-patterning combined with intertwined experiential development, which the interested reader may wish to study.

  • Corruption rampant in the oil industry, with taxpayers on the hook.
  • Why macroeconomics is different from regular economics.
  • Skidelsky dismantles Osborne.
  • On the internet, no one knows you're a stutterer. That is, until you podcast!
  • Tom Tomorrow draws a few crocodile tears.
  • Enlightenement, shmenlightenment, Texas style.
  • The media likes to be "balanced", except when it doesn't.
  • Bill Mitchell quote of the week: He advocates that the government directly employ excess (unemployed) labor at a sustainable minimum wage, rather than acceding to the business practice of casting them out onto the street.
"Given the overwhelming central bank focus on price stability, and the critical role of today’s buffer stocks of unemployed, we argue that functioning and effectiveness of the buffer stock is critical to its function as a price anchor.
Condition and liquidity are the keys. Just as soggy rotting wool is useless in a wool price stabilisation scheme, labour resources should be nurtured as human capital constitutes the essential investment in future growth and prosperity. There is overwhelming evidence that long-term unemployment generates costs far in excess of the lost output that is sacrificed every day the economy is away from full employment.
It is clear that the more employable are the unemployed the better the price anchor will function. The government has the power to ensure a high quality price anchor is in place and that continuous involvement in paid-work provides returns in the form of improved physical and mental health, more stable labour market behaviour, reduced burdens on the criminal justice system, more coherent family histories and useful output, if well managed."

No comments:

Post a Comment