Media reports into recent research have claimed that neuroscientists are now effectively able to perform ‘mind reading’. Such reporting inevitable raises ethical questions about what applications such research might eventually be put to, and, judging by some of the comments that the on-line versions of these articles have provoked, have alarmed some people regarding the eventual path that such research might take. But how accurate is the claim that neuroscientific techniques can read minds?
Early this year an article in the Guardian ( http://www.guardian.co.uk/science/2012/jan/31/mind-reading-program-brain-words ) reported that:
‘Scientists have picked up fragments of people’s thoughts by decoding the brain activity caused by words that they hear.’
Reporting on the same experiment the Daily Mail ( http://www.dailymail.co.uk/sciencetech/article-2095214/As-scientists-discover-translate-brainwaves-words–Could-machine-read-innermost-thoughts.html ) claimed:
‘It’s a staggering development that could have tremendous implications….judges could use mind-reading machines to find out if murder suspects are telling the truth….mind reading devices might be used to eavesdrop covertly on the most private thoughts and dreams.’
The experiment in question, conducted by Dr Brian Pasley and colleagues (1) involved the recruitment of patients who were to undergo brain surgery. The researchers placed electrodes upon the auditory areas of the brain during the period when the patients’ skulls were open and their cerebral cortex exposed. They then played the patients a sequence of different words and recorded the electrical activity generated by the auditory cortex in response to this speech. Using complex modeling procedures they were able to reconstruct the spoken words solely from the neural signals recorded by the electrodes. Furthermore they were able to successfully apply this model to the electrical responses generated by a separate set of words that had not been used in creation of the model (e.g. which were in effect ‘novel’ to the model) suggesting that the model could theoretically be applied to reconstruct any speech heard by the patient.
While these results are undoubtedly impressive, has the media coverage of them been accurate? In terms of the Guardian’s report, their claim that this represents a decoding of ‘fragments of thoughts’ seems to depend on a rather broad definition of the term ‘thoughts’. What the research did was to reconstruct auditory stimuli that the auditory cortex was in the process of analysing. What has been achieved therefore is the decoding, at a detailed level, of the perceptual process, NOT the reading of internally generated thoughts. This is a significant step away from ‘decoding thoughts’ as the process being decoded is entirely dependent on the presentation of an external stimulus. This doesn’t therefore represent ‘mind reading’ because the same result could theoretically be achieved without reference to the brain, e.g. by taking measurements from the relevant sensory organ or by just observing the sensory stimulus itself (2). Even if the research did represent mind reading, there seems little justification for the Daily Mail’s claim that the research could lead to ‘covert eavesdropping’. It should be obvious that the methodology required not only the opening up of the participant’s skull, but also the co-operation of the participant in allowing data to be taken for the construction of the model. Furthermore what is not mentioned by either article is that the reconstructed words were not actually intelligible to a human listener, but had to be ‘recognised’ via a speech recognition algorithm (an example of the reconstructed speech can be heard here: http://www.plosbiology.org/article/info%3Adoi%2F10.1371%2Fjournal.pbio.1001251#s5).
Actual Mind Reading?
While the results of Dr Pasley’s study required the participant’s brains to be exposed, other neuroimaging methods are not so intrusive, and could therefore be considered closer to the covert mind-reading reported by the Mail. Magnetic Resonance Imaging (MRI) allows brain activity to be measured in a non-invasive way, so that no surgery of any kind is required (although lying down in a scanner which costs millions of pounds and is the size of a small boat, is still required, making it far from ‘covert’!). MRI studies have produced some equivalent results to that of Pasley’s study, but using visual stimuli; with images (3) and short movies (4) having been reconstructed purely from data obtained from MRI scans. Of course such results don’t represent mind reading any more than Dr Pasley’s study, since they reflect a reconstruction of external sensory information. However other MRI studies have produced results that have allowed scientists to predict processes occurring within a participant’s brain that are not directly tied to the characteristic of external stimuli. A couple of studies by Yukiyasu Kamitani and Frank Tong (5,6) have shown that models can be created that allow an observer to identify to which stimulus a participant is (covertly) attending to. In effect these studies, and others like them, use the output from the perceptual processing mechanisms of the brain to identify how ‘top-down’ influences (such as expectation and attention) are driving perception. Strictly speaking they represent mindreading as although the mental processes in question are still involved in analysing external stimuli, it is not necessarily possible to garner the information provided by the MRI data in any other way (short of asking the person themselves). This is because the ‘top-down influences’ in question arise internally from the brain, rather than being a function of the external stimulus. Neuroimaging has enabled the concept of mind reading to be taken further however, into the realms of decoding mental events that don’t rely on any external stimulation at all. Recent studies have found that it is possible to decode what broad categories of objects someone is imagining, in the absence of any coincident external stimulation (7) although the performance level of the model is reasonably modest (~ 50%). Similarly, it also appears that the results of basic decision making processes can be identified from brain activity, with decisions relating to which button to press and when to press it (8) and whether a participant in lying (9) being decipherable using models constructed in a similar way to those already described. Interestingly the neural information that allows these decisions to be decoded occurs many seconds BEFORE the decision has actually been made, highlighting how conscious actions are likely driven by brain processes that are outside conscious awareness, rather than being the result of conscious ‘free will’. Most recently such work has been extended to more complex scenarios, with MRI data being used to predict at what point in solving an algebraic problem a child is at, and whether they are performing the calculation correctly (10).
The possibility of covert mind reading?
Clearly the aforementioned examples reflect mind reading, but do they represent the top of a ‘slippery slope’ that will lead to technology that will allow the sort of covert eavesdropping envisioned by the Daily Mail? The first impediment to such technology is the process of neuroimaging itself. MRI scanners are far from being portable enough to allow forced or covert application of brain scanning. Furthermore MRI scanning involves the production of a large magnetic field and the firing of electromagnetic pulses towards the object being imaged, both functions that would be totally impractical outside a controlled, isolated environment. Other neuroimaging methods, such as EEG, function by recording the electrical remnants of brain activity from outside the skull, and are therefore cheaper and more portable than MRI. However they lack the spatial resolution that would be required for any sophisticated mind reading application, and in any case they are extremely sensitive to external noise, again making them unsuitable for use outside of controlled environments.
Even if we assume that future technological advances would allow systems to be developed that would enable covert collection brain activity data, would such technology enable your innermost thoughts to be deciphered? There are a number of reasons to doubt that this would be possible. Current mind reading models are only able to distinguish between very broad categories of thoughts, or between very coarse categories of decisions (e.g. lie/truth, attending to one or other stimulus). To be able to read the specific details of an individual’s thoughts you would need models that distinguished between the literally billions of different things that someone could be thinking about, and the multitude of different decisions that they could make. To even create such models would involve the co-operation of individuals in a data collection process that would take an incalculable length of time. Even if such data were collected, and the subsequent required level of computation to create accurate models were possible, the ability to generalize such models to the brain activity of other individuals would rely on an assumption that every person’s brain being identical in terms of where different individual thoughts and memories are stored. This seems extremely unlikely, and is in fact counter to what we know about individual differences in brain anatomy and function. Thus while it is possible to aggregate data across participant to produce mind-reading for coarse decisions, it would be impossible to replicate such a method to distinguish between more subtle categories of thought. Even in situations where co-operation of the participant is attained, and only a coarse distinction between different psychological states is required, such mind reading techniques are problematic. Taking the example of the mooted ‘MRI Lie detector’ such a system will always be somewhat unreliable because, just like the current physiological lie detectors, they could be easily deceived if the participant can train themselves to act as if the truth is a lie (or vice versa). This is because the brain activity which is associated with lying most likely relates to the emotional and cognitive processes involved in creating a false story, rather than to lying per se. It follows that simply engaging in these same emotional and cognitive processes while telling the truth should produce neural activity which mimics that produced by a lie. If even the decoding of simple decisions can be subverted easily, it would seem impossible that attempts at more subtle discriminations of different thoughts would not be subject to even greater uncertainty. Finally it is important to note that all the forms of mind reading reviewed here are the result of probabilistic calculations. The parts of the brain that are deemed active at a certain point in time are the result of statistical computations as to whether a small signal is reflective of task-related neural activity or noise. Likewise the classification of such activity as belonging to one category of thought/decision over another is also based off probabilistic inference. There is no certainty in such a process; in fact it is fraught with uncertainty.
To conclude it seems very unlikely that neuroimaging methods will ever be able to perform the sort of mind reading predicted by scare stories in the press. In some cases such methods may not even represent a particular improvement on the sort of mind reading applications that already exist. What the mind reading research discussed in this article does allow is a greater understanding of how the brain works, which in turn provides insight into how the brain achieves the myriad feats it performs so frequently with apparent ease. The most fruitful practical application of such knowledge is likely to be in the treatment of patients with brain damage. For example the limited mind reading functions possible from existing neuroimaging methods may allow technology to be developed that would allow patients who suffer from brain damage to the extent that they cannot communicate using their peripheral nervous system, some primitive form of communication through their brain activity. In contrast your private thought and memories are likely to remain safe from the prying eyes of neuroscientists!
Image (top right) courtesy of Idea Go: http://www.freedigitalphotos.net/images/view_photog.php?photogid=809
(1) Pasley BN, David SV, Mesgarani N, Flinker A, Shamma SA, et al. (2012) Reconstructing Speech from Human Auditory Cortex. PLoS Biol 10(1): e1001251. doi:10.1371/journal.pbio.1001251 http://www.plosbiology.org/article/info%3Adoi%2F10.1371%2Fjournal.pbio.1001251
(2) Tong, F. & Pratte, M.S. (2012) Decoding Patterns of Human Brain Activity. Annual Review of Psychology, 63: 483-509. http://www.ncbi.nlm.nih.gov/pubmed/21943172
(3) Miyawaki, Y. Uchida, H. et al (2008) Visual Image Reconstruction from Human Brain Activity using a Combination of Multi-scale Local Image Decoders.. Neuron 60, 915–929, http://iopscience.iop.org/1742-6596/197/1/012021
(4) Nishimoto, S., Vu, A.T., et al (2011) Reconstructing Visual Experiences from Brain Activity Evoked by Natural Movies. Current Biology 21, 1641–1646 http://www.sciencedirect.com/science/article/pii/S0960982211009377
(5) Kamitani Y, Tong F. 2005. Decoding the visual and subjective contents of the human brain. Nat. Neurosci. 8:679–85 http://www.ncbi.nlm.nih.gov/pmc/articles/PMC1808230/
(6) Kamitani Y, Tong F. 2006. Decoding seen and attended motion directions from activity in the human visual cortex. Curr. Biol. 16:1096–102 http://www.ncbi.nlm.nih.gov/pmc/articles/PMC1635016/
(7) Reddy, L., Tsuchiya, N. & Serre, T. (2010). Reading the mind’s eye: Decoding category information during mental imagery. Neuroimage. 50(2) 818-825 http://www.ncbi.nlm.nih.gov/pmc/articles/PMC2823980/
(8) Soon CS, Brass M, Heinze HJ, Haynes JD. 2008. Unconscious determinants of free decisions in the human brain. Nat. Neurosci. 11:543–45 http://www.nature.com/neuro/journal/v11/n5/full/nn.2112.html
(9) Davatzikos C, Ruparel K, Fan Y, Shen DG, Acharyya M, et al. 2005. Classifying spatial patterns of brain activity with machine learning methods: application to lie detection. NeuroImage 28:663–68 http://www.sciencedirect.com/science/article/pii/S1053811905005914
(10) Anderson, J.R. (2012) Tracking Problem Solving by Multivariate Pattern Analysis and Hidden Markov Model algorithms. Neuropsychologia, 50(4) 487-498. http://www.sciencedirect.com/science/article/pii/S0028393211003605