The "hypercolumns" myth
Yehouda Harpaz
4 Mar 2002 [last updated 12 Nov 2004] 
related texts 

The "Hypercolumns" myth

The "hypercolumns" is one of the most stupid concepts in neuroscience. An "hypercolumn" (sometimes called "cortical module", also "ice-cube" model) is a square of 1-2 mm**2. The idea is that in the visual cortex this size of area is the smallest area that contains selectivity for all directions, from both eyes, and sometimes for all spatial frequencies. Therefore it is the "unit" or the "module" of the cortex.

That is plain nonsense. It would have made sense if we have a reason to assume that:

  1. The visual cortex is made of units or modules.
  2. Every unit or module is as small as possible.
  3. The kind of processing that a unit does requires a selectivity for all directions, and information from both eyes, and information from all spatial frequencies.
We don't have any evidence for any of this three assumptions, and I haven't seen anybody actually caliming that they have such evidence ([14Jan2002] Carpenter claims such modules with a very clear picture, but no evidence) . On the other hand, the visual cortex (like the rest of the cortex) is clearly a continuous sheet, and there is nothing in it that corresponds to the borders between the "hypercolumns". Thus the "hypercolumns" are completely fictitious concept.

In his Nobel Prize Lecture, Hubel actually mentions the fact that the hypercolumns borders are arbitrary, but it doesn't seem to bother him. In fact, he suggests that "the blocks of tissue may really be discrete", based on the finding of cytochrome oxydase "blobs". This is an example of "theory-driven blindness", because the hypercolmsn are clearly not discrete, and it was clear at the time of the lecture too.

Sometimes people claim that "hypercolumns" are an abstraction, but that is again nonsense. The features of the abstraction must corresponds to some features in the real thing to be an abstraction, and the "hypercolumn" concept does not correspond to anything.

The implicit thinking behind the "Hypercolumns" seem to be something like:

Both (2) and (3) are clearly invalid assumptions, but as long as they are not made explicit, it is difficult to see it. Since (4) is clearly false, one of (2) or (3) must be false.

As stupid as the "Hypercolumns" concept is, it is very pervasive. For example, if you search the net for hypercolumn and cortex, you will find many pages, and many of these are texts or outlines of lectures to psychology students. It seems that by now you are not likely to pass the exams in psychology unless you except the idea of hypercolumns. Here I list few examples.

Here you can see a figure of a "cortical module". Note that the borders of the "module" and the vertical lines dividing it to "orintation columns" and "spatial frequency columns" do not correspond to anything in the actual cortex. They are completely fictitious. The marking for left eye and right eye are less fictitious, because they correspond to where one eye become more dominant than the other eye, though the transition is not as sharp as the figure suggests.

This is another one. Look for "hypercolumns".

In this one(search for "ice-cube") it actually warns the reader not to take it literally, because the real "slabs" are not that regular. It doesn't tell the reader that vertical lines and the borders of the "ice-cube" are fictitious.

This one says explicitly that the hypercolumns "concerned with analyzing discrete area of visual field", which is obviously false, because the area is not discrete in any sense.

This one is written by a graduate student. He writes:

Striate cortex is composed of many hypercolumns arranged in a retinotopic map of the visual world.
Clearly in this guy's mind the hypercolumns are independent entities.

Another one. It is not obvious if this one thinks that the hypercolumns are independent entities or not.

This one repeats the normal nonsense, and also gives explicit number for the difference between orientation columns (10 degrees), which strengthen the impression that they are really separate.

This lecture summary makes explicit the broken logic that leads to the "hypercolmn" being a "elementary computational unit":

The elementary computational unit of cortex is the hypercolumn [fig 23-17]
- all the circuitry to analyze 1 pt. on retina
360 degrees of orientation
both eyes
several blobs
- lateral connections link similar columns of different hypercolumns
For a change in this review (Lund et al, Cerebral Cortex, Vol. 13, No. 1, 15-24, January 2003) they say:
There is, however, no fixed boundary to such hypercolumns as there is a continuous change in property and mean visual field position across the cortex.
But they leave it to the reader to realize that this means that the concept of hypercolumns is fiction, rather than saying it themselves.