Current Biology 17, 323–328, February 20, 2007 ª2007 Elsevier Ltd All rights reserved DOI 10.1016/j.cub.2006.11.072
Reading Hidden Intentions
in the Human Brain
John-Dylan Haynes,1,2,3,4,5,* Katsuyuki Sakai,6
Geraint Rees,4,5Sam Gilbert,4Chris Frith,5
and Richard E. Passingham5,7
1Max Planck Institute for Human Cognitive
and Brain Sciences
2Bernstein Center for Computational Neuroscience
3Charite ´ – Universita ¨tsmedizin
4Institute of Cognitive Neuroscience
University College London
WC1N 3AR London
5Wellcome Department of Imaging Neuroscience
Institute of Neurology
University College London
WC1N 3BG London
6Department of Cognitive Neuroscience
Graduate School of Medicine
University of Tokyo
7Department of Experimental Psychology
University of Oxford
OX1 3UD Oxford
When humans are engaged in goal-related process-
ing, activity in prefrontal cortex is increased [1, 2].
However, it has remained unclear whether this pre-
frontal activity encodes a subject’s current intention
. Instead, increased levels of activity could reflect
preparation of motor responses [4, 5], holding in mind
a set of potential choices , tracking the memory of
previous responses , or general processes related
to establishing a new task set. Here we study subjects
who freely decided which of two tasks to perform and
covertly held onto an intention during a variable delay.
Only after this delay did they perform the chosen task
strate that during the delay, it is possible to decode
from activity in medial and lateral regions of prefrontal
cortex which of two tasks the subjects were covertly
intending to perform. This suggests that covert goals
can be represented by distributed patterns of activity
in the prefrontal cortex, thereby providing a potential
neural substrate for prospective memory [8–10].
During task execution, most information could be de-
coded from a more posterior region of prefrontal cor-
tex, suggesting that different brain regions encode
goals during task preparation and task execution. De-
coding of intentions was most robust from the medial
prefrontal cortex, which is consistent with a specific
role of this region when subjects reflect on their own
We directly addressed whether the current intentions of
a subject were encoded in specific regions of prefrontal
cortex. This was achieved by assessing whether multi-
variate pattern recognition  could be used to decode
that subject’s covert intention from activity patterns in
prefrontal cortex. If a cortical region indeed represents
a current intention, it must have some way of encoding
a set of different potential goals. One possibility is that
it uses a spatial code, with different, spatially segre-
gated neural subpopulations encoding different inten-
tions. Unfortunately, because of the limited spatial reso-
lution of human neuroimaging, most researchers have
restricted their analyses to activity averaged across ex-
tended regions of cortex. This leaves unclear whether
there are any regions encoding intentions in a spatially
distributed fashion. However, it has recently emerged
to study fine-grained neural representations, even when
they are encoded at a finer scale than the resolution of
the measurement grid [12, 13]. This technique is power-
fulenough torevealdistributed representations ofvisual
images in occipital and temporal brain areas [11–14].
In order to investigate whether a subject’s current in-
tentions are reflected in such distributed response pat-
terns in prefrontal cortex, we required subjects to freely
select what task they wished to perform. Specifically,
they chose either adding or subtracting two numbers
(Figure 1). After the subject had freely decided upon
one of the two tasks, there was a variable delay of be-
rial (two numbers) was presented. The variable delay
rendered the onset of the task-relevant material unpre-
dictable and thus required the subject to maintain
a high state of preparation even across long intervals
. Shortly after the two numbers, a response screen
was presented that contained four numbers: one was
the correct answer for addition, one was the correct an-
swer for subtraction, and the two other numbers were
similar but incorrect numbers. Subjects only rarely
chose one of the two incorrect numbers (average 5%),
indicating that they were correctly performing the task
and not responding randomly. From the choice of one
task the subject had chosen for the current trial. How-
ever, it is important to note that there was no way of tell-
ing which task the subject had freely selected prior to
the response, because there was no explicit instruction
and no behavioral response prior to the onset of the re-
sponse screen, and subjects responded randomly (see
Figure S1 in the Supplemental Data available online).
Also, because the arrangement of numbers on the re-
sponse screen was random, there was no possibility
for the subject to prepare their motor response. This en-
sured that any information we could decode from brain
activity during the delay was not related to covert prep-
aration of motor responses [4, 5].
We recorded brain responses with functional mag-
netic resonance imaging at 3 Tesla while subjects were
performing the free-selection task. In order to investi-
gate which cortical regions encode the subject’s current
intention, we next assessed whether it was possible to
decode from the spatial pattern of signals in each local
region of the brain which intention the subject was
covertly maintaining [11–14]. For this, we applied multi-
variate pattern recognition to spatial patterns of brain
responsesunder thetwopossibleintentions (seeExper-
imental Procedures and Figure 2 for details on this anal-
ysis). We found that indeed several regions predicted
whether the subject was currently covertly intending to
perform the addition or subtraction task (Figure 2). The
highest decoding accuracy of 71% was achieved in me-
dial prefrontal cortex (T= 4.62, p = 0.001, see Figure 2,
‘‘MPFCa’’). Importantly, however, decoding in this re-
gionwasnot possibleduring taskexecution, suggesting
that the intention was encoded in this brain region only
during the delay and not during task execution. In con-
dial wall was not informative during the delay, but only
during the execution of the freely chosen task (Figure 2,
‘‘MPFCp’’). Besides medial prefrontal cortex, there were
also several regions of lateral prefrontal cortex where
decoding accuracy was lower, but still above chance
level (Figure 2). Also in these regions, decoding was at
chance level during task execution. Interestingly, only
a region of anterior-medial prefrontal cortex showed
an overall increase of activity during the delay period
while subjects had covertly formed a decision but
were still waiting to execute the task (Figure S2). As in
previous studies [10, 15], the duration of increased neu-
ral activity corresponded to the delay in the current task,
with longer delays leading to longer fMRI responses.
However, this region with an overall signal increase
was more anterior to the region that encoded the sub-
ject’s intentions. Importantly, there was no difference
between the two intentions in the overall level of activity
(T= 20.46; p = 0.67) in medial prefrontal cortex, sug-
gesting that the intentions were not encoded in different
global levels of activity but in the detailed spatial pat-
terns of cortical responses.
To summarize, we have demonstrated that regions of
both medial and lateral prefrontal cortex contain localiz-
able task-specific representations of freely chosen in-
tentions. In accordance with our findings, activity in sev-
eral regions of human prefrontal cortex (including the
frontopolar, lateral, medial, and prefrontal cortex) is in-
creased during diverse executive processes such as at-
tending to and thinking about intentions [16, 17], task-
switching [18–20], set-shifting , multitasking ,
storing goals over a delay period [9, 10, 15, 23], branch-
ing and processing of subgoals [24, 25], and free task
selection . However, these previous studies left un-
codes signals that are specific for the current task. In-
creased levels of activity during task preparation might
instead reflect unspecific preparatory signals, such as
maintaining a representation of the set of all potential
choices , tracking the memory of previous responses
, or general preparation. Our new findings resolve this
crucial question by showing for the first time that pre-
frontal cortex encodes information that is specific to
a task currently being prepared by a subject, as would
be required for regions encoding a subject’s intentions.
In accordance with our findings, single cells in monkey
lateral prefrontal cortex can prospectively encode ex-
ual increase in the information about simple saccadic
movement sequences while animals learn to perform
a sequence correctly . Cells have also been reported
in the same area that code for specific moves while the
monkey is waiting to move a cursor so as to negotiate
a maze . Here we show that in humans, a network
of brain regions, including not only lateral but also me-
dial prefrontalcortex, contains such task-specific repre-
Although intention-related information was encoded
in both lateral and medial regions of prefrontal cortex,
decoding accuracy was highest in the medial region.
Figure 1. Delayed Intention Task
At the beginning of each trial, the word ‘‘se-
lect’’ was presented that instructed the sub-
jects to freely and covertly choose one of
two possible tasks, addition or subtraction.
After a delay during which subjects covertly
maintained their intention, two numbers
were presented and subjects were then re-
quired to perform the selected task (addition
or subtraction) on the two numbers. A re-
sponse screen then appeared showing two
correct answers (for either addition or sub-
traction) and two incorrect answers. Subjects
pressed a button to indicate which answer
was correct for the task they had performed.
From the button press, it was possible to de-
termine the covert intention of the subject
during the previous delay period.
One possible explanation may be that the current study
allowed subjects to freely select which task to perform,
whereas in most previous studies the task goal was
specified by the experimenter. Medial prefrontal cortex
is especially involved in the initiation of willed move-
ments and their protection against interference . In-
creased levels of medial activation are also found when
task sets have to be internally generated as opposed to
being fullyexternally cued[26,31]andinsimilarcasesof
underdetermination [32–34]. For example, a direct com-
parison between voluntary task selection and externally
eral, prefrontal regions  during the free selection of
tasks. Medial frontal cortex is also activated when sub-
jects reflect on their own mental states [35, 36]. Interest-
ingly, we also observed a division of labor between pos-
terior and anterior regions of medial prefrontal cortex,
tion whereas the posterior regions encode goals during
tions byprevious authorsthat there isan anterior/poste-
rior gradient on the medial frontal surface . Similarly,
the area at which activity is enhanced when subjects at-
tend to their intention to perform a simple finger move-
ment is more anterior than the area at which activity is
enhanced when they attend to the finger movement
itself [17, 34].
An interesting question for future research is the de-
gree to which the encoding in different prefrontal areas
reflects sustained maintenance of intentions across
multiple trials, or trial-by-trial switching of intentions.
Because of the random trial-by-trial alternation of sub-
jects in our study, we measured predominantly shift-tri-
als (see Figure S1), and therefore we are unable to ad-
dress the differences between sustained and transient
encoding. Previous studies have reported evidence for
both sustained  and shift-related  activity in me-
dial prefrontal cortex, so it would be interesting to apply
similar decoding-based techniques to cued paradigms
where the number of switch and stay trials can be better
An important feature ofour paradigmisthatit ensured
that subjects could not covertly prepare for a specific
movement prior to the onset of the response-mapping
screen. This is in contrast to previous studies on pro-
spective coding that used event-related potentials.
These studies have shown that it is possible to decode
on a single trial whether subjects are going to choose
to move the left or right fingers  and that this informa-
tion is present even prior to the time at which the sub-
jects believe themselves to be making a decision .
However, these signals are recorded over motor-related
brain regions and thus are likely to reflect the covert
preparation of specific motor programs immediately
preceding the execution of a movement [4, 5]. In our
study, we can rule out the possibility that decoding dur-
ing the delay was based on motor preparation. Please
note that because the task-relevant stimuli and the re-
sponse screen occurred in tight temporal sequence,
we are unable to separate motor preparation from en-
coding of intentions during the execution period. An in-
teresting question is whether one could decode which
Figure 2. Brain Regions Encoding the Subjects’ Specific Intentions during Either Delay or Execution Periods
In order to search in an unbiased fashion for informative voxels, we used a ‘‘searchlight’’ approach , which examines the information in the
local spatial patterns surrounding each voxel vi.
Left: A spherical searchlight centered on one voxel (vi) was used to define a local neighborhood. For each scanning run, the spatial response
pattern in this local spherical cluster was extracted during preparation of either subtraction or addition. We then trained a pattern classifier
with a subset of the data to recognize the typical response patterns associated with covert preparation of the two mathematical operations
(see Experimental Procedures) and measured the local decoding accuracy. Then, the searchlight was shifted to the next spatial location.
Middle: Highlighted in green are medial brain regions (superimposed on a saggital slice of an anatomical template image) where this local clas-
sifier was able to decode significantly above chance which intention the subjects were covertly holding in an independent test data set. High-
lighted in red are regions where it was possible to decode the intention during the execution of the task.
Right: Decoding accuracy in searchlight locations with above-chance decoding during the delay period (MPFCa, anterior medial prefrontal cor-
tex [MNI 3,42,15]; MPFCp, posterior medial prefrontal cortex [MNI 11,32,38]; LLFPC, left lateral frontopolar cortex [MNI 236,54,12]; LIFS, left
inferior frontal sulcus [MNI 22,36,12]; RMFG, right middle frontal gyrus [MNI 48,24,45]; LFO, left frontal operculum [MNI 239,9,9]; error bars in-
dicate SEM). In the anterior medial prefrontal cortex (MPFCa), decoding during the delay (green bars) was highest but was at chance level during
the task execution (red bars) after onset of the task-relevant stimuli. In contrast, a more posterior and superior brain region on medial prefrontal
cortex (MPFCp) encoded the chosen task only once it had entered the stage of execution, but not during the delay period. Several other regions
of lateral prefrontal cortex also encoded information during the delay, but not during the execution period.
Reading Hidden Intentions
pacing of individual trials, we were not able to reliably
analyze the period prior to the cued time of selection.
However, an important implication of our study is that
in future it might be possible to use decoding to reveal
which specific brain areas unconsciously determine
the intention that a subject is about to choose .
Importantly, we found that overall delay-related activ-
ity in prefrontal cortex was indistinguishable under both
conditions. There was no evidence, therefore, that pre-
paring to perform one task was more difficult than pre-
paring to perform the other. Furthermore, this finding
means that the two intentions are encoded, not by
some increase in global activity, but by different spatial
response patterns. This raises the intriguing question of
the precise neural basis of these cortical patterns en-
coding different intentions, given that there is a strong
overlap between cortical responses to different tasks
. One possible explanation may be that cells in spe-
cific regions of prefrontal cortex have a functional spe-
cialization for either of the two tasks, and that there is
a fine-grained clustering of cells with similar properties
that is smaller than the size of conventional areas. For
example,in visual cortex, information encoded insimilar
fine-grained patterns of visual cortex can be read out by
a‘‘biased sampling’’ or ‘‘aliasing’’ offine-grained feature
columns by the individual fMRI voxels [11–13] and is
confirmed by simulations based on realistic neural to-
pographies (see Supplemental Data in ). This raises
the question whether the informative spatial patterns
nar architecture in prefrontal cortex, where cells might
be clustered according to similar roles in selective cog-
nitive control. Such a columnar architecture has been
highly debated as a general principle of cortical organi-
zation [40, 41] and has been claimed for the prefrontal
cortex . Alternatively, our sampling patterns might
reflect the sampling of a distributed population code
for different tasks as has been proposed from the find-
ings of similar studies on object recognition . Future
optical imaging studies will be able to extend our find-
ings by studying the local spatial topography of execu-
tive signals in prefrontal cortex. An important question
for future studies will be whether the medial prefrontal
cortex is generally involved in encoding specific tasks
during intentional choices or whether encoding in this
region is specific for tasks such as the preparation of
addition and subtraction.
Taken together, our results extend previous studies
on the processing of goals in prefrontal cortex in several
important ways. They reveal for the first time that spatial
response patterns in medial and lateral prefrontal cortex
encode a subject’s covert intentions in a highly specific
fashion. They also demonstrate a functional separation
in medial prefrontal cortex, where more anterior regions
encode the intention prior to its execution and more
posterior regions encode the intention during task exe-
cution. These findings have important implications not
only for the neural models of executive control, but
also for technical and clinical applications, such as the
further development of brain-computer interfaces, that
might now be able to decode intentions that go beyond
simple movements and extend to high-level cognitive
Participants and Experimental Design
Three male and five female subjects (age between 21 and 35) gave
written informed consent to participate in the experiment, which
was approved by the ethics committee at the Institute of Neurology,
University College London. All subjects were right-handed and had
normal or corrected to normal visual acuity.
at fixation that instructed the subjects to rapidly select one of the
lay of between 2.7 and 10.8 s, during which the subject was in-
structed to prepare for the task. Because of the variable delay, the
onset time of the task-relevant stimuli was not predictable, requiring
extended delay [10, 15]. Then, the task-relevant stimuli were pre-
sented, which consisted of two 2-digit numbers presented above
and below the fixation spot. Subjects were instructed to either add
or subtract the two numbers in accordance with the task they had
previously covertly chosen. Then after 2 s, a ‘‘response-mapping’’
screen was presented that showed four numbers, one in each visual
quadrant on the screen. Two of these numbers were correct re-
sponses (one for addition and one for subtraction) and two were in-
correct responses. Subjects responded with one of four response
buttons operated by the left and right index and middle fingers.
The keys corresponded to the positions of the four numbers on
ysis (see below) was performed on signals related to brain activity
prior to onset of the screen with task stimuli and thus 2 s before
the response assignment, so decoding could not have been based
on covert motor preparation because the mapping of correct and
incorrect responses to keys was randomized from trial to trial. The
distribution of phase durations during the main experiment (i.e.,
sequences of N trials where subjects chose the same task) followed
an exponential distribution, as would be assumed if subjects chose
randomly on each trial which task to perform (Figure S1). Prior to the
experiment, subjects practised the task for 7 min. During each
scanning run, subjects performed 32 trials.
A Siemens Allegra 3T scanner with standard head coil was used to
acquire functional MRI volumes (42 slices, TR = 2730 ms, resolution
3 3 3 3 1.5 mm3). For each subject, 8 runs of functional MRI data
were acquired each with 155 images. To avoid susceptibility arte-
facts, slices were tilted 20?and the resolution in read-out direction
was increased to 1.5 mm. The first three images of each run were
discarded to allow for magnetic saturation effects.
ThefMRIdata weremotion corrected,spatiallynormalizedtoastan-
dard stereotaxic space (Montreal Neurological Institute EPI tem-
plate), and resampled to an isotropic spatial resolution of 3 3 3 3
3 mm3in SPM2 (http://www.fil.ion.ucl.ac.uk/spm). The first analysis
was designed to identify brain regions where activity was signifi-
cantlyincreased during thedelayperiodwhilesubjectswereprepar-
ing for the task (Figure S2). This analysis was performed with a gen-
eral linear model as implemented in SPM2. The model consisted of
four boxcar regressors, each convolved with a canonical haemody-
namic response function. Each regressor modelled either the delay
or execution period of one of the two task types. Prior to the GLM
analysis, the data were smoothed with a Gaussian kernel of 6 mm
FWHM to account for the anatomical variability across subjects
and to satisfy the assumptions of Gaussian random field theory .
The second analysis was designed to identify regions where spa-
tially distributed fMRI activation patterns carried information about
general linear model as above but now based on unsmoothed data.
This change was made to maximize sensitivity and allow extraction
of the full information present in the spatial patterns of prefrontal
cortex, which would have been reduced by the smoothing. Then, in
order to search in an unbiased fashion for informative voxels, we
used anovel variant of the‘‘searchlight’’approach ,which exam-
ines the information in the local spatial patterns surrounding each
voxel vi(seeFigure 2,left).Thus, foreach vi,weinvestigated whether
its local environment contained spatial information that would allow
decoding of the current intention.
For a given voxel vi, we first defined a small spherical cluster of N
voxels c1.Nwith radius of three voxels centered on vi. For each
voxel c1.Nin the fixed local cluster, we extracted the unsmoothed
parameter estimates for delay-period activity separately for covert
preparation of the addition and the selection task. This yielded two
N-dimensional pattern vectors xr,1.Nand yr,1..Nfor each run r, repre-
senting the spatial response patterns in the local cluster during co-
vert preparation for addition and subtraction. Next, we used multi-
variate pattern recognition to assess how much intention-related
information was encoded in the local pattern. To achieve this, we
assigned the pattern vectors xr,1.Nand yr,1..Nfor seven of the eight
imaging runs to a ‘‘training’’ data set that was used to train a linear
support vector pattern classifier  (with fixed regularisation pa-
rameter C = 1) to correctly identify response patterns related to
the two different intentions the subject was currently holding. The
classification was performed with the LIBSVM implementation
The amount of intention-related information present within this lo-
cal cluster could then be assessed by examining how well the inten-
tions during the remaining independent eighth or ‘‘test’’ data set
were classified. Good classification implies that the local cluster of
voxels spatially encodes information about the specific current in-
tention of the subject. In total, the training and test procedure was
repeated eight times, each with a different run assigned as test data
set, yielding an average decoding accuracy in the local environment
of the central voxel vi(8-fold crossvalidation). Then, the procedure
was repeated for the next spatial position at voxel vj. The average
decoding accuracy for each voxel was then used to create a
3-dimensional spatial map of decoding accuracy for each position
viin prefrontal cortex. Because the subjects’ images had previously
beennormalizedtoacommonstereotactictemplate, itwas possible
to extend previous local decoding approaches  and perform
a second-level analysis where we computed on a voxel-by-voxel
basis how well decoding could be performed on average across
all subjects from each position in the brain. This yielded a spatial
map of average decoding accuracy that is plotted in green in
Figure 2. We also performed a similar pattern classification with the
parameter estimates for task execution as opposed to the delay
period. This is plotted in Figure 2 in red.
Two Supplemental Figures can be found with this article online at
This work was supported by the Max Planck Society, the Wellcome
Trust, and the Mind-Science Foundation. The authors would like to
thank M. Brass and M. Ullsperger for valuable comments.
Received: September 11, 2006
Revised: November 24, 2006
Accepted: November 27, 2006
Published online: February 8, 2007
1. Frith, C., Gallagher, H., and Maguire, E.A. (2004). Mechanisms
of control. In Human Brain Function, R.S.J. Frackowiack, ed.
(London: Elsevier), pp. 329–364.
2. Rowe, J.B., Stephan, K.E., Friston, K., Frackowiak, R.S., and
Passingham, R.E. (2005). The prefrontal cortex shows context-
specific changes in effective connectivity to motor or visual
cortex during the selection of action or colour. Cereb. Cortex
3. Passingham, R.E., and Lau, H.C. (2006). Free choice in the hu-
man brain. In Does Consciousness Cause Behavior? S. Pockett,
4. Haggard, P., and Eimer, M. (1999). On the relation between brain
potentials and the awareness of voluntary movements. Exp.
Brain Res. 126, 128–133.
5. Blankertz, B., Dornhege, G., Schafer, C., Krepki, R., Kohlmor-
gen, J., Muller, K.R., Kunzmann, V., Losch, F., and Curio, G.
(2003). Boosting bit rates and error detection for the classifica-
tion of fast-paced motor commands based on single-trial EEG
analysis. IEEE Trans. Neural Syst. Rehabil. Eng. 11, 127–131.
6. Frith, C.D. (2000). The role of dorsolateral prefrontal cortex in the
selection of action as revealed by functional imaging. In Control
of Cognitive Processes: Attention and Performance XVIII, S.
Monsell and J. Driver, eds. (Cambridge, MA: MIT Press), pp.
7. Hadland, K.A., Rushworth, M.F., Passingham, R.F., Jahanshahi,
M., and Rothwell, J.C. (2001). Interference with performance of
a response selection task that has no working memory compo-
nent: an rTMS comparison of the dorsolateral prefrontal and
medial cortex. J. Cogn. Neurosci. 13, 1097–1108.
8. Miller, E.K., and Cohen, J.D. (2001). An integrative theory of
prefrontal cortex function. Annu. Rev. Neurosci. 24, 167–202.
9. Burgess, P.W., Scott, S.K., and Frith, C.D. (2003). The role of the
rostral frontal cortex (area 10) in prospective memory: a lateral
versus medial dissociation. Neuropsychologia 41, 906–918.
10. Sakai, K., and Passingham, R.E. (2003). Prefrontal interactions
reflect future task operations. Nat. Neurosci. 6, 75–81.
11. Haynes, J.D., and Rees, G. (2006). Decoding mental states from
brain activity in humans. Nat. Rev. Neurosci. 7, 523–534.
12. Haynes, J.D., and Rees, G. (2005). Predicting the orientation of
invisible stimuli from activity in human primary visual cortex.
Nat. Neurosci. 8, 686–691.
13. Kamitani, Y., and Tong, F. (2005). Decoding the visual and sub-
jective contents of the human brain. Nat. Neurosci. 8, 679–685.
14. Haxby, J.V., Gobbini, M.I., Furey, M.L., Ishai, A., Schouten, J.L.,
and Pietrini, P. (2001). Distributed and overlapping representa-
tions of faces and objects in ventral temporal cortex. Science
15. Sakai, K., and Passingham, R.E. (2006). Prefrontal set activity
predicts rule-specific neural processing during subsequent
cognitive performance. J. Neurosci. 26, 1211–1218.
16. Den Ouden, H.E., Frith, U., Frith, C., and Blakemore, S.J. (2005).
Thinking about intentions. Neuroimage 28, 787–796.
17. Lau, H.C., Rogers, R.D., Haggard, P., and Passingham, R.E.
(2004). Attention to intention. Science 303, 1208–1210.
18. Dove, A., Pollmann, S., Schubert, T., Wiggins, C.J., and von Cra-
mon, D.Y. (2000). Prefrontal cortex activation in task switching:
an event-related fMRI study. Brain Res. Cogn. Brain Res. 9,
19. Brass, M., and von Cramon, D.Y. (2002). The role of the frontal
cortex in task preparation. Cereb. Cortex 12, 908–914.
20. Braver, T.S., Reynolds, J.R., and Donaldson, D.I. (2003). Neural
mechanisms of transient and sustained cognitive control during
task switching. Neuron 39, 713–726.
21. Konishi, S., Nakajima, K., Uchida, I., Kameyama, M., Nakahara,
K., Sekihara, K., and Miyashita, Y. (1998). Transient activation
of inferior prefrontal cortex during cognitive set shifting. Nat.
Neurosci. 1, 80–84.
22. Burgess, P.W., Veitch, E., de Lacy Costello, A., and Shallice, T.
(2000). The cognitive and neuroanatomical correlates of multi-
tasking. Neuropsychologia 38, 848–863.
23. Burgess, P.W., Quayle, A., and Frith, C.D. (2001). Brain regions
involved in prospective memory as determined by positron
emission tomography. Neuropsychologia 39, 545–555.
24. Koechlin, E., Basso, G., Pietrini, P., Panzer, S., and Grafman, J.
(1999). The role of the anterior prefrontal cortex in human cogni-
tion. Nature 399, 148–151.
25. Braver, T.S., and Bongiolatti, S.R. (2002). The role of frontopolar
cortex in subgoal processing during working memory. Neuro-
image 15, 523–536.
26. Forstmann, B.U., Brass, M., Koch, I., and von Cramon, D.Y.
(2006). Voluntary selection of task sets revealed by functional
magnetic resonance imaging. J. Cogn. Neurosci. 18, 388–398.
Reading Hidden Intentions
27. Rainer, G., Rao, S.C., and Miller, E.K. (1999). Prospective coding Download full-text
for objects in primate prefrontal cortex. J. Neurosci. 19, 5493–
28. Averbeck, B.B., Sohn, J.W., and Lee, D. (2006). Activity in pre-
frontal cortex during dynamic selection of action sequences.
Nat. Neurosci. 9, 276–282.
29. Mushiake, H., Saito, N., Sakamoto, K., Itoyama, Y., and Tanji, J.
(2006). Activity in the lateral prefrontal cortex reflects multiple
steps of future events in action plans. Neuron 50, 631–641.
30. Paus, T. (2001). Primate anterior cingulate cortex: when motor
control, drive and cognition interface. Nat. Rev. Neurosci. 2,
31. Forstmann, B.U., Brass, M., Koch, I., and von Cramon, D.Y.
(2005). Internally generated and directly cued task sets: an in-
vestigation with fMRI. Neuropsychologia 43, 941–952.
32. Botvinick, M.M., Cohen, J.D., and Carter, C.S. (2004). Conflict
monitoring and anterior cingulated: an update. Trends Cogn.
Sci. 8, 539–546.
33. Walton, M.E., Devlin, J.T., and Rushworth, M.F.S. (2004). Inter-
actions between decision making and performance monitoring
within prefrontal cortex. Nat. Neurosci. 7, 1259–1265.
34. Lau, H.C., Rogers, R.D., and Passingham, R.E. (2006). On mea-
suring the perceived onsets of spontaneous actions. J. Neuro-
sci. 26, 7265–7271.
35. Amodio, D.M., and Frith, C.D. (2006). Meeting of minds: the
medial prefrontal cortex and social cognition. Nat. Rev. Neuro-
sci. 7, 268–277.
36. Northoff, G., Heinzel, A., de Greck, M., Bernpohl, F., Dobro-
wolny, H., and Panksepp, J. (2006). Self-referential processing
in our brain—a meta-analysis of imaging studies of the self.
Neuroimage 31, 440–457.
37. Weidner, R., Pollmann, S., Mu ¨ller, H.J., and von Cramon, D.Y.
(2002). Top-down controlled visual dimension weighting: an
event-related fMRI study. Cereb. Cortex 12, 318–328.
38. Libet, B., Gleason, C.A., Wright, E.W., and Pearl, D.K. (1983).
Time of conscious intention to act in relation to onset of cerebral
activity (readiness-potential). The unconscious initiation of
a freely voluntary act. Brain 106, 623–642.
39. Duncan, J., and Owen, A.M. (2000). Common regions of the
human frontal lobe recruited by diverse cognitive demands.
Trends Neurosci. 23, 475–483.
40. Mountcastle, V.B. (1997). The columnar organization of the neo-
cortex. Brain 120, 701–722.
41. Horton, J.C., and Adams, D.L. (2005). The cortical column:
a structure without a function. Philos. Trans. R. Soc. Lond.
B Biol. Sci. 360, 837–862.
42. Constantinidis, C., Franowicz, M.N., and Goldman-Rakic, P.S.
(2001). Coding specificity in cortical microcircuits: a multiple-
electrode analysis of primate prefrontal cortex. J. Neurosci.
43. Worsley, K.J., Marrett, S., Neelin, P., Vandal, A.C., Friston, K.J.,
and Evans, A.C. (1996). A unified statistical approach for deter-
mining significant signals in images of cerebral activation.
Hum. Brain Mapp. 4, 58–73.
44. Kriegeskorte, N., Goebel, R., and Bandettini, P. (2006). Informa-
tion-based functional brain mapping. Proc. Natl. Acad. Sci. USA
45. M} uller, K.R., Mika, S., Ra ¨tsch, G., Tsuda, K., and Scho ¨lkopf, B.
(2001). An introduction to kernel-based learning algorithms.
IEEE Trans. Neural Netw. 12, 181–202.