I’ve come to the mind-reading issue at an interesting time. Recently, Nikos Logothetis published an article called What we can do and what we cannot do with fMRI. It’s a fascinating paper, and numerous blogs picked up on it. The best summary comes from Mind Hacks and is ominously entitled The great fMRI smackdown cometh.
About the application of pattern-classification techniques (e.g. MVPA) to fMRI data, Logothetis says the following:
In humans, fMRI is used routinely not just to study sensory processing or control of action, but also to draw provocative conclusions about the neural mechanisms of cognitive capacities, ranging from recognition and memory to pondering ethical dilemmas. Its popular fascination is reflected in countless articles in the press speculating on potential applications, and seeming to indicate that with fMRI we can read minds better than direct tests of behaviour itself. Unsurprisingly, criticism has been just as vigorous, both among scientists and the public. In fact, fMRI is not and will never be a mind reader, as some of the proponents of decoding-based methods suggest, nor is it a worthless and non-informative ‘neophrenology’ that is condemned to fail, as has been occasionally argued.
A lot of this has to do with how one defines “mind-reading,” and whether or not what’s being read is low-level (e.g. the orientation of a visual stimulus) or high level (e.g. the presence of a mental state). There are numerous examples of papers which exploit fMRI data to erroneously support what I would call very high-level conclusions about human behavior, such as whether or not subjects have anxiety regarding Mitt Romney. (This and many other such studies are handily refuted by The Neurocritic, who also had some interesting commentary on Logothetis’s paper.) However, I think the study by Kay et al. on the identification of novel, real-world images suggests that, at lest for low-level, sensory information, fMRI might be quite the mind-reader indeed. There’s a good editorial in Science by Greg Miller which has, I think, a pretty balanced view of things, resorting to neither abject pessimism nor fanciful speculation.
The risk I think Logothetis and others are pointing out is that inferring a mental state from brain imaging data puts one on pretty shakey ground. I think this is especially true for studies which don’t take into account distributed patterns of brain activation, but even pattern-classification based studies need to be very careful about making the leap to state attribution. Just because we can say, in a sense, that there is a hierarchy inside a rhesus macaque’s brain that differentiates between cats and dogs, we can’t necessarily say that the monkey knows what a dog is and that it has a degree of similarity to a cat equal to X. It’s possible that the observed hierarchical organization might be completely unconscious.
Logothetis gets specific about his reservations regarding pattern classification techniques:
Such multivariate analyses or pattern-classification-based techniques (decoding techniques) can often detect small differences between two task or stimulus conditions—differences that are not picked up by conventional univariate methods. However, this is not equivalent to saying that they unequivocally reveal the neural mechanisms underlying the activation patterns.
And he’s right. But the strength of pattern classification of fMRI data is not that it can reveal underlying neural mechanisms, nor that it can simply detect small differences between patterns of activation (cheating past the spatial resolution limit, basically), but that it allows us to evaluate the similarity of patterns of activiation based on a variety of arbitrary criterea. It’s not cool because it gets us closer to some underlying machinery, it’s cool because it lets us talk about similarity and difference of patterns of activation in a quantitative way. This has surprising power.
References: