David Marr was brilliant, but wrong

David Marr was a brilliant Neuroscientist who died too young, in 1980, at the age of 35. Marr’s work was theoretical — he was at the leading edge of a computational wave.* Marr’s contributions spanned many areas of Neuroscience: cerebellum, hippocampus, and especially vision. Marr is also well known for proposing that brain/behavior function should be approached in three phases that are largely sequential:

  1. The computational level: what is the problem that confronts the animal?
  2. The algorithmic level: How is it logically solved? (including shortcuts)
  3. The implementation level: How does the brain do it?

This seems to make a great deal of sense. For example, how can you study how the brain solves a problem before you know what the problem is? A specific example comes from my field of study: single cell recording in the hippocampus of behaving rats. Scientists have been able to record the action potentials from single cells since about 1940, and have been able to do this in behaving animals since about 1960. Let’s imagine you are a Neuroscientist in 1960, interested in the hippocampus. You have the ability to record from single cells, you are interested in the hippocampus, but you have no idea what the hippocampus does for a living. You put electrodes into the hippocampus, listen to the firing of a single cell and hope to get clues. If you listened to David Marr, he would likely say you were crazy.

times squareOne analogy I’ve heard often is to the old scrolling news board in Times Square**. Let’s imagine you are a visitor from Mars, and you want to know what this flashing board is about. You behave like a Neuroscientist: you replace a single light bulb with a voltmeter and record the “firing” of the bulb. You could sit there a long time without learning anything, even if you were a very smart Martian.

place cell

Firing map of a hippocampal place cell. Recorded in cylinder for 15 minutes. Dark pixels are non-zero firing rates while rat’s head in pixel. Yellow = 0 rate.

Well this is essentially what John O’Keefe did. He put single cell recording electrodes into the hippocampus, listened to the firing of a single cell, and tried to figure out the code. The perhaps surprising result is that he figured it out. Hippocampal neurons are place cells. Each place cell will fire action potentials in a restricted region of space, called the “firing field”. By listening to the discharges of individual neurons and watching the rat’s behavior, he figured this out. Notice that O’Keefe started at Marr’s stage 3, without a grasp of the answers from stages 1 and 2. The result of O’Keefe’s discovery has been 35 years of productive work on the function of the hippocampus as a cognitive mapping system.


John O'Keefe

John O’Keefe

Was O’Keefe lucky? Of course. Nature was helpful. There are many neurons in the brain (and the hippocampus) where the code is not so clear. But others, following O’Keefe’s approach (but with hints of what to look for) discovered head-direction cells, theta cells, grid cells and conjunctive cells in the hippocampal formation. Although O’Keefe was “lucky” it’s hard to imagine another route led to the same end  point.

The take home point, I think, is that scientific strategy is largely an intuitive affair. This is especially true for large advances. We in the hippocampal field are indebted to John O’Keefe, but we don’t know how he did it. He might not know, either. But he wasn’t following the advice of David Marr.

*Are there conceptual paradigm shifts in Neuroscience? I believe we are on the midst of one: understanding the brain as a computational engine, but far removed from a computer.
** I head this analogy frequently from Jim Ranck, my post-doctoral advisor.


2 thoughts on “David Marr was brilliant, but wrong

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s