Slashdot Mirror
Face Recognition Needs 3 Areas Of Human Brain
sushant_bhatia_progr writes "Nature has an article on the recent discovery that face recognition in humans targets 3 areas of the human brain. Using mugshots of celebrities, Pia Rotshtein at University College London and her colleagues have shown that there are at least three separate areas for processing and recognising faces. One processes the physical features of the face, one decides whether or not the face is known, and a third retrieves information about that person, such as their name. Rothstein's team used a computer to create a series of images in which the countenance of film star Marilyn Monroe gradually morphed into that of former British prime minister Margaret Thatcher, or that of James Bond actor Pierce Brosnan transformed into current prime minister Tony Blair."
Haven't you worked it out yet? It doesn't matter where they come from, All politicians are crooks.
True it is no surprise that the three intuitive components of face recognition (see, recognize, identify) show activation in different regions of the brain. But these type of "it's obvious & intuitive" comments follow many scientific discoveries, especially those in psychology, and entirely miss the point of the experimental method - to prove (or disprove) those intuitions.
The problem with such a racist is not their thoughtcrime, failing to recognize racial differences, but their actions. If they can't (or won't) notice differences among individuals of other races, that's they're problem. When they burn these people's houses down, beat them in nightclubs, refuse to hire them, or do other bad things, it doesn't really matter that their facial recognition is wired wrong.
When we make thoughts illegal, we're faced with legislating people's minds. Not only politically catastrophic in a free society, but probably medically irresponsible to pretend we are in control of all the results. We have a flawed, but much more successful, history of managing behavior. We should stick to what we know until we've improved it to adequacy, before messing with minds and all the worse consequences at stake.
--
make install -not war
the countenance of film star Marilyn Monroe gradually morphed into that of former British prime minister Margaret Thatcher
You don't need computers for that. You just need to wake up next to someone you don't remember meeting.
For more information on the subject, listen to the song "9 Coronas".
I am officially gone from
Modularization: Great for OO programming, crappy for the human brain.
IAWMUHTIPORI (I am writing my undergraduate honors thesis in philosophy on related issues) What sort of "modularization" are you referring to? Modularization of peripheral systems (input/output systems, i.e., the senses)? If so, you must realize that you would be in the extreme minority in opposing a modular architecture for these systems (see Jerry Fodor's Modularity of Mind, the standard treatment on peripheral systems modularity with which the vast majority of cognitive scientists agree).
If you are talking about central cognitive systems (belief formation, inference to the best explanation, theory of mind, etc.) things get a bit more complicated. Recent empirical evidence seems to indicate that anatomical modularization of central systems is probably not thoroughgoing in the human brain. However, a lack of any real anatomical modularity does not mean that the human brain is not ultimately modular, in some sense of the word.
The best evidence for conceptual modularity (that is easy for the non-expert to understand) is implicit in the arguments against the other major alternative for cognitive architecture: distributed connectionism (e.g., Parallel Distributed Processing). Specifically, distributed connectionist networks may be able to do certain specialized tasks -- such as optical character recognition -- rather well. But it is next to impossible to get a distributed connectionist network to do more than one thing well without the system eventually grinding to a halt. This is, in part, the result of the inability of a truly distributed connectionist network to maintain a manageable search space when serving multiple purposes.
A modular central architecture, in contrast, can do any number of distinct tasks without the sort of combinatorial explosion that a distributed connectionist architecture is apt to run up against. This is because the modules within a modular central architecture are thought to be highly specialized to handle specific tasks. This feature of modular systems also allows us to see how the brain develops and might have evolved -- one specialized system at a time (for the most part). It is extremely difficult to even imagine how a general problem solver, such as a distributed connectionist network, develops or could have evolved.
The most significant problem for modular cognitive (central) systems, then, doesn't involve a lack of thoroughgoing anatomical modularization, since we are often not talking anatomical modules when we talk about modularity. The main problem for the type of modularity that is popular these days has to do with the lack of a good way to tie all of the modules together to make a flexible system that has the surface appearance of being a general problem solver (as the anti-adaptationist Fodor points-out in his most recent book, The Mind Doesn't Work that Way , which is primarily a criticism of Steven Pinker's popular How the Mind Works ).
In the past couple of years, several theories have been put forth to explain modular integration. Perhaps the most notable among these is that the natural language module serves as the modular integrator. The original article in which this theory was articulated in detail has been made available by the author on his website. The article with criticisms and the author's response to the criticisms is available only in the print edition of the journal Behavioral and Brain Sciences ("The Cognitive Functions of Language" in Volume 25, Issue 6).
Again, then, the issue is a good bit more complex than the parent post indicated. In fact, if the cur
I am curious as to what you think the conclusions of these debates in the "non-philosophical areas of cognitive science" have been. I could cite numerious articles that have come from people in cognitive science outside of philosophy in the past five years defending positions all across the board, from massive modularity to distributed connectionism and everything in between. Just look at the article from BBS that I mentioned above, particularly the replies, the response to the replies, and the associated citations in the bibliography of that article. It seems to me that the individual perception as to the current status of the debate depends on what area of cognitive science the individual works in.
People who work in AI seem to take modularity for granted, currently, so they think that the debate is over and modularity has won. People in linguistics seem to like distributed connectionism a bit more than people in AI, although they are not generally sold on it. Psychologists are either agnostic or split on the issue, depending on whether or not they think the evolutionary approach has anything to offer their field. Neuroscientists apparently often buy into the connectionism more closely resembles the actual brain line, and so the majority of them still work with PDP-like models, but only when they don't have regular access to real brains and fMRI or the like. Philosophers, of course, are open to some possibilities that people in each of the other constituent disciplines of cognitive science see as being silly, although some sort of modularity seems to be winning out among philosophers (except for at UCSD).
But my overall impression is that in no case except maybe for AI do most people consider the debate completely passe. Taking into consideration your "moist eyes" comment and your perception that the debate is passe, I'm tempted to believe that you were, at the very least, trained in AI. Is that right?