Here's a cool article in Haaretz that gives a nice overview of various views regarding neuroimaging (PET, fMRI) studies and their relative scientific value.

I thought the section on "reading minds" was pretty interesting. It starts with this:

Prof. Chris Frith, a brain researcher from University College, London, one of the letter's organizers, told Haaretz Magazine in a telephone interview that it later turned out that a number of the researchers who took part in the study have ties to a company that markets brain-imaging machines. He assails the attempt to seek this type of explanation for complex social phenomena, and notes that the public is very attracted to the idea of "reading minds." One sees the images of the brain with the colored regions and thinks one is seeing what happens to people in their thoughts, which is something no one was able to do before, Frith explains. But what people are not told is that these colors are simply numbers, representing levels of blood flow or electrical activity. There is no way to look at these images and know what a person is thinking.

Now it's important to be skeptical about any and all lines of research, but I liked the line "simply numbers". It's kind of a red flag when a scientist is saying data is "simply numbers".

There's a kind of back and forth in the article after that:

The notion that it will be possible to compile in the future some sort of "brain dictionary" containing all human thoughts or feelings is baseless, Frith believes. One can identify a few very simple activities that are carried out in the brain, he notes, such as face recognition, but there is a tremendous difference between that and compiling a dictionary per se.

In contrast, Sompolinsky maintains explicitly that producing a lexicon of this sort is feasible: "If we image your brain and examine activity in regions when you say the word 'apple' or when you see an apple - there is no reason to think that there is no specific region that is connected to that object. And when you dream, I can know that if this pattern arises in your brain, you were dreaming about an apple. It is still not clear whether the pattern will be the same for other people, too. But in my opinion, the differences are far smaller than we may believe."

Prof. Marcelo Dascal, a philosopher who teaches at Tel Aviv University (TAU) and an international expert on the psychology of language, decries this approach. "Anyone who thinks that nouns exist somewhere in the brain has no idea of what awareness is, how we think and what language is. The idea that every time we think of a certain word, a fixed area of the brain is activated is simply idiotic. The whole greatness of human language lies in the fact that we can use the same word for a 1,001 different purposes."

Ah...simply idiotic. Well, just because you can use a given word in 1,001 different contexts doesn't mean that there's still not a particular part of your brain that is reliably active when you use that word.

Of particular interest is this brand new study published in Nature, Identifying natural images from human brain activity by Kay et al.

Two subjects were shown 120 different pictures (10 times each) of natural scenes and fMRIs were taken of their brain activity for each viewing. Based on the fMRI data alone, their computer algorithm was able to achieve 92 and 72 percent accuracy for each subject respectively in identifying which object they were looking at. Basically this means that a computer analyzing a brain scan could reliably predict whether a human was looking at an apple or a horse.

When the number of pictures was increased to 1,000, the computer still achieved an accuracy over 80%. That seems pretty impressive to me, though I'd like to hear some criticism of the research.

I'd also like to hear a better defense of the view expressed by Dascal above. I understand the view that our brain states are fluid and unlikely to globally repeat the same pattern, since we never exactly experience the same thing twice. But we do experience extremely similar stimuli over and over as part of different global experiences. It would be very strange indeed if there weren't a particular area that becomes active every time we see something like an apple. If a different part of our brain was active every time we saw the same object, that would mean that the encoding is basically arbitrary. If this is really what Dascal would argue, then that seems idiotic, but maybe I'm misunderstanding him.

Anyway, it's interesting stuff. If you don't know much about neuroimaging, go have a look at the Haaretz article.