How David Marr might approach Neuroscience
The current discourse surrounding "neuroscience" often strikes me as a rather circuitous path, lacking the fundamental clarity we require for true understanding. Before we speak of neurons, synapses, or even these so-called "neural networks" that seem to capture the popular imagination, we must first address the *computational theory*. What is the goal of this intricate biological system, the brain? What problems is it trying to solve, and why are these particular solutions, honed by evolution, appropriate?
We are too often distracted by the hardware. The rich tapestry of biological matter, the firing of electrical impulses – these are crucial, yes, but they are merely the *implementation*. The truly important thing for progress, for insight, is the algorithm. What are the representations that the brain constructs, and what are the algorithms it uses to manipulate them, to transform raw sensory input into meaningful perception and action?
Consider vision. It is not simply about light hitting the retina. It is about recovering the structure of the three-dimensional world from two-dimensional projections. This is a fundamentally ill-posed problem, requiring sophisticated computational solutions. We need to understand the steps involved: the extraction of basic visual primitives like edges and surfaces, their integration into a coherent viewer-centered representation, and the subsequent derivation of a more object-centered, stable description.
If this field, "neuroscience," is to truly advance, it must adopt this tripartite approach. We need to move beyond mere correlation – observing brain activity and trying to assign it a function. We must rigorously define the computational problems the brain is designed to solve, devise elegant and efficient algorithmic…
Imagined perspective — an AI synthesis grounded in David Marr’s recorded ideas and methods, not a quotation or a statement they actually made.