Wednesday, August 22, 2007

How many daddies do I have?

Imagine the problem facing a newborn baby. The baby has a strong bond with the mother, and is, from very early on, used to her presence, touch, voice, smell, and body.

But not the father.

Each day, "daddy" appears in a new, very different form. One day he's bearded and smelling like the beer he had 20 minutes ago. Another day he's clean-shaved and with a completely distinct smell. His voice changes from day to day, from loud singing and playing and laughing, to a "normal tone", to low whispers. Now, the task for a baby--or better, for a baby's brain, is to answer "how many daddies do I have"?

Let me rephrase: given the immense variety of incoming stimuli, how can a baby find out that all that different mass of information originates from the same stable thing in the cosmos named "daddy"?

Welcome to the representation invariance problem.

Every single moment of your life you are bombarded with a huge amount of information. Your eyes alone have enough bandwidth leading information to your brain to account for an enormous number of phone calls. A million nerve cells provide visual information to your brain. 30 thousand nerve cells provide the input to "hearing". And the thing is, after these initial cells fire up the information inside the brain, things change. The image that your eyes receive rapidly ceases to be an image inside the brain, and becomes a pattern of neural firings so complex no neuroscientist has mapped it in detail. After some steps, the pattern of firings cannot be correlated to the original projected image anymore--your brain is working on something entirely different, a representation, a guess of what's out there in the world.

The brain--or better, the mind--sees the world. But it is dark, completely, absolutely, dark, inside your brain. The representation does not feel dark, but it is made in utter darkness.

Seeing is creating an interpretation of what's in the world. What kind of stable things are out there that could be generating this particular input images? This is what the brain does: under complete darkness, a pattern of firings become a lively mental image. Maybe it's "daddy", maybe it's "mommy", maybe it's a toy, maybe something else entirely.

And the original image is so huge, so immense, that in all probability, in your whole lifetime, it is most likely you will never see the exact same image projected to the exact same retinal cells. So your brain is receiving a gigantic flux of constantly changing information, and it has to provide, for your understanding of the world, a representation of what is out there. The staggering thing is: this representation does not vary easily. It is an invariant representation. You see the same person under a new light, or a new angle, and you never think that suddenly it's another person--for the representation is invariant. Now here's a quote from "On Intelligence":

The problem of understanding how your cortex forms invariant representations remains one of the biggest mysteries in all of science. How difficult, you ask? So much that no one, not even using the most powerful computers in the world, has been able to solve it. And it isn't for lack of trying. (Hawkins, 2004, p.78)
I think Mr Hawkins isn't accurate in this one. We do have a solution, "available, today"--but very few people know about it. This is the ace up our sleeve; our upper hand in the race to find the organizing principles of cognitive technology.